Book a Call
Mario Alexandre — AI Systems Engineer
Engineering Consultant — DLux Digital

AI engineered like critical infrastructure.
Documented, monitored, owned.

Production AI systems, built to engineering standards. You own the code.

BSEE · University of South Florida 7 yrs Electrical · Luanda, Angola Tampa, FL EN / PT / ES
Browse Services Book a Call
Built On
Claude API OpenAI Python FastAPI n8n Stripe PostgreSQL AWS ElevenLabs Whisper vLLM MCP
Background

Engineering is engineering. The scope just keeps getting bigger.

Before I wrote my first line of Python, I spent seven years designing electrical systems for commercial buildings in Luanda, Angola — the kind of work where a mistake means a building goes dark. That taught me how to think about systems: redundancy, fault tolerance, making things work under constraints most engineers never face.

When I came to the US to formalize that knowledge with a BSEE from the University of South Florida, I discovered the same engineering discipline applies to an entirely different kind of system. Software. AI. Automation. The medium is different — code instead of copper — but the method is identical: architect the system, build it to production grade, make it reliable enough that your business can depend on it.

Today I run DLux Digital, building production AI systems for companies whose operations need to work without constant oversight. Real money, real clients, no prototypes. Every system documented, monitored, and handed over — you own it.

7+
Years EE — Angola
83K+
Lines Production Python
500+
Transcripts Processed
99%
Pipeline Reliability
40
Production Services

Industries I Build For

// Where I've Shipped

Every industry has its own friction profile. These are the verticals where I've shipped production systems and understand the operational realities — billing models, regulatory edges, common stack, where the labor really goes.

// 01

Coaches & Course Creators

Content extraction from Zoom recordings, gated course delivery, payment-to-CRM bridges. Real example: client with 55+ hrs/mo of unclipped coaching gold.

// 02

Real Estate & Hospitality

Lead qualification, dynamic pricing tooling, calculator pages, Airbnb operations automation. Built for high-volume listing operators.

// 03

Agencies & Service Firms

Outbound sales automation, AI proposal generation, sales call analyzers. Replaces SDR seats and lifts close rates.

// 04

Government & Enterprise

Multilingual deployments (EN/PT/ES), document intelligence at scale, compliance-grade audit trails. Lusophone-Africa specialty.

// 05

SaaS & Technology

Custom MCP servers, self-hosted LLM stacks, AI knowledge agents. For teams who need their AI to actually do things — not just describe them.

// 06

Regulated Industries

Healthcare, fintech, legal: GDPR/HIPAA/SOC 2 monitoring, audit-ready logs, on-prem LLM deployment. Engineering-grade compliance.

Core Automation Services

01 – 05 / 40
Small Business Essentials
01

AI Voice
Receptionist

Answers every call 24/7. Books appointments. Handles customer questions without humans. Custom voice cloning makes it sound like your business.

Dentists, salons, law firms, clinics, HVAC. Clients paying $3K+/mo for a front desk that still puts callers on hold.

$3K–$5K build
+ $300–$500/mo
2–4 hours
Retell AI / VapiElevenLabsGoogle Calendar API
02

AI Review
Manager

Monitors every Google review. Auto-responds to positive ones. Flags negatives for the owner. Sends review requests to happy customers.

Restaurants, gyms, chiropractors, auto shops. Any business whose revenue depends on their Google rating.

$500–$1,500 build
+ $200–$500/mo
1–2 hours
n8nGoogle Business APIClaude API
03

AI Lead
Qualifier

Website chatbot live 24/7. Engages every visitor. Asks qualifying questions, captures contact info, books serious buyers onto the calendar.

Real estate agents, coaches, consultants, agencies. Anyone losing leads at night because nobody's there to respond.

$5K–$8K build
+ $500–$1K/mo
4–8 hours
Voiceflow / BotpressClaude APICalendlyHubSpot
04

AI Ad Creative
Pipeline

Scrapes competitor ads daily. Identifies winning hooks. Generates 50–100 ad variations with scripts, images, and voiceovers, in an afternoon instead of a month.

Ecom brands, DTC companies, marketing agencies paying $5K–$15K/mo to creative agencies for the same output.

$1K–$3K/mo
per brand
One weekend
ClaudeMidjourneyRunwayElevenLabsCron job
05

AI Sales Call Analyzer

Auto-transcribes every sales call. Analyzes for objections, winning phrases, and drop-off points. Grades each rep's performance. Pushes a coaching summary to Slack or the CRM.

B2B sales teams, agencies, real estate brokerages, car dealerships. Any business that knows bad calls cost money but can't review every recording.

$2K–$4K build + $300–$800/mo
4–8 hours
Whisper / FirefliesClaude APIn8nSlack / CRM integration

Engineering-Grade Services

06 – 12 / 40
Senior Engineer Differentiators
06

AI Document
Intelligence

Extracts, classifies, and routes data from invoices, contracts, intake forms, and PDFs. Reduces manual document review by 80%+ and eliminates transcription errors at scale.

Legal firms, insurance companies, healthcare practices, logistics. Any operation drowning in unstructured documents.

$8K–$20K build
+ $500–$1.5K/mo
1–2 weeks
Claude APIAWS Textractn8n / MakePostgreSQL
07

Predictive
Analytics

Custom ML pipelines that forecast churn, inventory demand, revenue, or fraud from your actual business data. Dashboards that tell you what's going to happen, not just what happened.

E-commerce, SaaS, fintech, and any data-rich business whose decisions are still based on last month's spreadsheet.

$10K–$30K build
+ $1K–$2K/mo
2–4 weeks
Python / scikit-learnMetabase / SupersetdbtSnowflake / BigQuery
08

AI Compliance & Audit Automation

Continuous monitoring for GDPR, HIPAA, SOC 2, or CCPA. Logs every action. Flags violations in real time. Generates audit-ready reports on demand. Built for regulated industries where a manual audit costs $50K+.

Healthcare, fintech, legal, HR tech. Any business facing regulatory scrutiny that handles compliance manually.

$15K–$40K build + $2K–$5K/mo
3–6 weeks
PythonClaude APIDatadog / SplunkPostgreSQLCustom audit APIs
09

Edge AI
Deployment

On-device LLM and ML inference for Jetson, Raspberry Pi, NPU, and mobile. Privacy-first, zero API cost after deployment, runs offline. Hardware-aware quantization and thermal/power budgeting.

Privacy-sensitive industries, factory floors, vehicles, retail kiosks. Anywhere data shouldn't leave the device or connectivity is unreliable.

$8K–$25K build
+ $300–$800/mo
2–5 weeks
ONNX / TFLiteLlama.cpp / vLLMJetson / RPiQuantization (INT4/INT8)
10

Custom MCP
AI Tool Servers

Build Model Context Protocol servers. Give Claude, GPT, or any LLM authenticated access to private tools, internal databases, and proprietary APIs. Your AI gets real hands instead of just hallucinated answers.

SaaS companies, internal AI initiatives, product teams who want their LLM to actually do things, not just describe doing them.

$4K–$10K build
+ $300–$700/mo
1–2 weeks
MCP SDK (TypeScript / Python)FastAPISSE / WebSocketOAuth / JWT
11

Self-Hosted
LLM Stack

Private Ollama or vLLM deployment on client GPU (or rented Hostinger / RunPod / Hetzner). Zero per-token cost, zero data leakage, full model selection: Qwen, Llama, Mistral, DeepSeek. Includes nginx, auth, monitoring.

Companies hitting 5-figure monthly OpenAI bills. Regulated industries that can't send data to third parties. Anyone hitting rate limits.

$6K–$18K build
+ $400–$1K/mo
1–3 weeks
vLLM / OllamaQwen / Llama / Mistralnginx + CaddyPrometheus / Grafana
12

AI System Audit & Reliability Engineering

When an existing AI system "works in demos but fails in production," I run a BSEE-grade audit. The audit covers monitoring coverage, error budgets, edge-case behavior, drift detection, fallback paths, and cost analysis. Deliverable is a written report with prioritized fixes, plus an SOW for the rebuild if you want it done.

Companies with existing AI systems that mysteriously break, hallucinate in production, blow past budgets, or work in dev but not in scale. Every "AI agency" ships demos. Few engineer for production.

$3K–$8K flat audit
(rebuild SOW separate)
3–7 days
Written reliability report · Failure-mode catalog · Monitoring gap analysis · Remediation roadmap with effort estimates

Growth & Operations

13 – 21 / 40
High ROI · Fast to Ship
13

Internal AI
Knowledge Agent

A company's internal AI that answers employee questions using their actual docs, SOPs, Notion pages, and Confluence. No more pinging colleagues for answers that are already written down somewhere.

Mid-size companies with growing teams, onboarding friction, and institutional knowledge locked in siloed docs.

$5K–$12K build
+ $400–$800/mo
3–5 days
Claude APIPinecone / WeaviateNotion / Confluence APISlack bot
14

AI Proposal
Generator

Client fills out a qualification form. AI builds a customized, branded proposal with scope, pricing, and timeline, delivered in minutes instead of hours. Closes deals faster.

Agencies, contractors, consultants, architects, IT service firms. Anyone spending 2–4 hours per proposal on copy-paste work.

$3K–$6K build
+ $200–$400/mo
1–3 days
Claude APITypeform / TallyGoogle Docs APIn8n
15

Content Extraction
Pipeline

Your best content is rotting on a hard drive. AI watches every Zoom recording, podcast, or workshop. Identifies the moments that stop a scroll: emotional peaks, framework drops, contrarian takes. Clips them, formats them, queues them for posting. Your content calendar fills itself from material that already exists.

Coaches, founders, podcasters, creators sitting on hours of recorded gold. Anyone whose bottleneck is extraction, not creation.

$4K–$10K build
+ $500–$1.5K/mo
1–2 weeks
WhisperClaude (semantic scoring)ffmpegBuffer / Hypefury / Make
16

SEO Article
Pipeline

Automated content factory: keyword research → article generation with Schema.org structured data → editorial pass → publish to CMS → submit to Google Search Console. SEO-graded output at scale, not slop.

Brands losing organic traffic to faster competitors. Agencies servicing content clients. Anyone whose blog hasn't grown in 6 months.

$5K–$15K build
+ $500–$2K/mo
1–3 weeks
Claude / GPTSchema.org JSON-LDWebflow / WordPress / CaddyGSC APICloudinary
17

Payment & Access
Control Bridge

Stripe / ThriveCart / Teachable / Shopify → CRM with token-gated content delivery. Customer pays, gets a unique access token, content unlocks automatically. Webhook-driven, recoverable, audited. Built for products where access leakage costs money.

Course creators, paid newsletters, gated SaaS, info-product businesses doing $30K+/mo whose access flow is held together with tape.

$4K–$10K build
+ $300–$700/mo
1–2 weeks
Stripe / ThriveCart webhooksFastAPIJWT tokensPostgreSQLCRM API
18

Outbound Sales
Automation

Automated prospect discovery (LinkedIn, Apollo, web scraping), ICP qualification with LLM scoring, personalized message generation, multi-step follow-up sequences. Replaces a $5K/mo SDR with a system that runs 24/7 and never gets tired.

B2B sales teams, agencies, consultants. Any business whose pipeline is bottlenecked by outbound capacity.

$5K–$12K build
+ $500–$1.5K/mo
1–2 weeks
Apollo / Sales NavigatorClaude (personalization)Smartlead / InstantlyCRM API
19

AI Avatar &
Video Pipeline

Branded talking-head videos generated from script. Voice cloning + face animation + B-roll + captions. From script to publish-ready MP4 in under an hour. Build it once, generate hundreds.

Course creators, marketers, agencies, language-localization teams. Any operation that publishes video at volume but burns hours per piece.

$3K–$8K build
+ $400–$1K/mo per brand
3–7 days
HeyGen / D-ID / AkoolElevenLabsRemotion / ffmpegWhisper (captions)
20

Custom Browser
Automation

Production-grade web automation: scraping behind logins, multi-step form filling, internal-tool RPA, anti-bot evasion, CAPTCHAs handled. Headed and headless modes, persistent sessions, residential proxy rotation. Built for jobs no zapier-grade tool can handle.

Operations teams stuck in manual data entry. Agencies with one-off scraping needs. Businesses dependent on legacy tools without APIs.

$4K–$10K build
+ $400–$1K/mo
1–2 weeks
Playwright / PuppeteerClaude (vision parsing)BrightData / SmartproxyQueue + retry infra
21

Telegram Bot Infrastructure

Business-grade Telegram bots: payment handling, user authentication, AI-powered replies, file uploads, group management, broadcast queues, audit logs. Built on FastAPI + python-telegram-bot. Self-hosted, no Botpress lock-in, no message-credit metering.

Communities, premium subscriber groups, customer-support deployments. Anyone using Telegram as a business channel and outgrowing free bot builders.

$2K–$6K build + $200–$600/mo
3–7 days
python-telegram-botFastAPIStripe / crypto paymentsClaude APIPostgreSQL

Hardware + Energy + Industrial AI

23 – 28 / 40
AI Value × EE Foundation — Production Systems, Not Prototypes
23

AI-Driven Smart
Building Energy

Reinforcement-learning controllers on HVAC loops, deep-learning load forecasting per zone, ML demand-response optimization against live utility tariffs. Combines BACnet / Modbus building-automation expertise with production ML pipelines — an AI that actually commands equipment, not just dashboards it. Grounded in USF Control Systems + Power Systems + Energy Delivery coursework + feedback-control concept depth. Typical bill reduction: 15–30%.

Commercial properties with 4+ digit monthly bills and peak-demand charges — offices, hotels, retail chains, multi-tenant buildings, hospitals, data centers.

$12K–$35K build
+ $800–$2K/mo
3–6 weeks
BACnet / ModbusPython MLInfluxDB / GrafanaUtility API integrations
24

AI-Driven Industrial
Predictive Maintenance

Deep-learning anomaly detection on raw sensor streams — fused with classical DSP features (FFT spectra, wavelet decomposition, bearing-signature analysis). Detects bearing wear, insulation breakdown, harmonic distortion, current imbalance before failure. Classical vibration analysis on its own misses novel failure modes; pure ML on its own misses physics. The fusion catches both. Grounded in USF Digital Signal Processing + Power Lab + Electromechanical coursework + concepts: Fourier Transform (101 refs), convolution, windowing.

Manufacturing plants, utilities, data centers, building operators with critical electrical equipment. Anywhere unplanned downtime costs more than $10K/hour.

$15K–$40K build
+ $1.5K–$3K/mo
4–8 weeks
Vibration / current sensorsPython (scikit + PyTorch)FFT / wavelet analysisSCADA integration
25

AI-Assisted Solar PV
& Microgrid Design

LLM + optimization engine ingests utility bills, irradiance data, and building plans → automated load-profile analysis, optimal panel/inverter sizing under cost and production constraints, ML-based tariff forecasting for payback modeling, and permit-ready single-line diagrams. Combines generative AI reasoning over tariff docs with classical power-system sizing math. Grounded in USF Power Systems + Power Electronics + Energy Delivery coursework. From sales call to engineered proposal in hours, not weeks.

Solar installers, EPC firms, microgrid developers, commercial property owners. Replaces 2–4 days of senior-engineer time per proposal.

$8K–$20K build
+ $400–$800/mo
2–4 weeks
PVLib / NREL SAMMATLABAutoCAD / Revit APIsUtility tariff data
26

Embedded AI &
TinyML Firmware

Quantized neural networks running on microcontrollers (ESP32, STM32, RP2040). TinyML wake-word detection, sensor fusion, on-device anomaly classification — with bare-metal interrupt handling, RTOS scheduling, and battery-aware duty-cycling underneath. AI that respects microsecond deadlines and 100-mW power budgets. Grounded in USF Embedded Systems + C Programming + Digital Systems + embedded-rtos concept (Real-Time/Embedded is the #3 most-cited concept in Mario's coursework). Distinct from Edge AI (#09) — this is silicon-class, not Jetson-class.

IoT product companies, hardware startups, OEMs adding "smart" features without adding cellular bills or cloud latency.

$15K–$45K build
+ $500–$1K/mo
3–8 weeks
C / C++ / Rust embeddedTFLite Micro / ExecuTorchPlatformIOFreeRTOS / Zephyr
27

AI × DSP —
Signal Intelligence

Where classical DSP meets deep learning. Custom neural filters for non-stationary noise, ML-augmented beamforming, learned demodulators, deep-learning denoising on biosignals. Fourier / wavelet / FIR / IIR fused with CNN / transformer feature extraction — neither alone beats the hybrid. Public proof: the sinc-LLM project itself applies Nyquist-Shannon sampling theorem to LLM prompts — that IS this service, demonstrated on AI. Grounded in USF Digital Signal Processing + Signal Processing + Communications coursework + concepts: convolution, sampling-theorem, windowing, z-transform, Modulation (top concept, 121 source refs).

Audio engineering firms, medical-device startups, RF/comms teams, industrial vibration-monitoring vendors. Anyone with "we have noisy signal data" that no pure-DSP or pure-ML consultant has solved.

$200–$400/hr
or $5K–$15K flat
1–4 weeks
MATLAB / SimulinkPython (NumPy / SciPy)GNU RadioCustom FIR/IIR/wavelet
28

AI-Powered Engineering Document Automation

Multimodal LLMs (Claude Vision, GPT-4V) ingest architectural plans and equipment schedules → outputs single-line diagrams, panel schedules, load schedules, O&M manuals, and NEC / IEC code-compliance reports. The AI does the generation; classical EE rule-engines do the code-compliance verification. Generates documentation that takes a junior engineer 2–6 weeks per project, in hours. Grounded in USF EE Design + Circuit Analysis coursework + 7 years of commercial-building electrical design in Luanda — I know what reviewers reject and why, and I've encoded that into the rule engine.

Electrical engineering firms, MEP consultancies, design-build contractors, EPC firms. Anywhere drafters and EITs are bottlenecked on documentation while PEs are the constraint on stamp.

$10K–$25K build + $500–$1.2K/mo
2–5 weeks
Claude (vision)AutoCAD / Revit APIsPythonNEC / IEEE rule enginePDF generation

Hardware × AI: Specialized Engineering Domains

29 – 34 / 40
35 USF EE Courses × Production AI — Services Nobody Else Can Credibly Offer
29

ML-Augmented
RF & SDR Systems

Neural-network signal classification on SDR streams, ML-based interference cancellation, AI-driven spectrum sensing, autonomous protocol identification. Combines classical RF physics (link budgets, antenna design, modulation) with deep-learning signal intelligence — a fusion no pure-RF firm or pure-ML team can deliver alone. Grounded in USF coursework: RF/Microwave + Wireless Circuits + Wireless Mobile Security + concept depth in Modulation (121 source refs).

SDR product companies, defense / SIGINT contractors, IoT vendors needing self-healing spectrum use, satellite / aerospace startups. Anyone whose RF problem became ML-shaped the moment they collected enough data.

$15K–$50K build
+ $800–$2K/mo
3–8 weeks
CST / HFSS / ADSGNU Radio / SDRLoRa / Zigbee / BLEVNA / spectrum analyzer
30

AI-Optimized Grid
& Power Electronics

ML for load forecasting, transformer-failure prediction, dynamic line rating, inverter health monitoring, BESS state-of-health estimation. Combines power-systems engineering (PSS/E, fault analysis, IEC 61850) with production ML pipelines that turn telemetry into operational decisions. Grounded in 4 USF courses: Power Systems + Power Electronics + Power Lab + Energy Delivery.

Utilities running aging infrastructure, IPPs optimizing dispatch, EV / BESS integrators predicting cell degradation, industrial plants tracking transformer health. Anyone whose SCADA collects more data than humans can read.

$15K–$50K build
+ $1K–$2.5K/mo
3–8 weeks
PSS/E · ETAP · DigSILENTPSCAD · PLECSMATLAB / Simulink PowerIEC 61850 protocols
31

AI-Enhanced Radar
& Detection Systems

Deep learning fused with classical radar signal processing — neural-network target classification on top of matched-filter detections, ML clutter rejection, learned CFAR thresholds, autonomous track initiation. Outperforms classical-only or ML-only systems by exploiting both prior physics and learned patterns. Grounded in USF Radar Systems coursework + concepts: matched-filter, detection-theory, radar-tracking — paired with production deep-learning ops.

Automotive ADAS suppliers (FMCW + ML), drone / UAV defense, perimeter security, weather and remote-sensing instrumentation. Anyone whose detection problem is "classical works but misses 5% — that 5% is the lawsuit."

$20K–$60K build
+ $1K–$3K/mo
4–10 weeks
MATLAB Phased Array ToolboxPython (NumPy / SciPy)CFAR detectionKalman / particle tracking
32

AI-Powered Wireless
& IoT Security

LLM-driven autonomous penetration-testing agents, ML-based wireless intrusion detection, AI traffic-pattern anomaly classifiers on cellular / BLE / Wi-Fi, automated firmware vulnerability triage. Combines wireless-protocol expertise with agentic security tooling — your AI keeps testing while your humans sleep. Grounded in USF Wireless Mobile Security coursework + network-protocols concept depth.

IoT product companies needing continuous (not annual) pen-testing, blue teams overwhelmed by alert volume, critical-infrastructure operators, regulators auditing connected products.

$8K–$25K audit
or $2K–$5K/mo retainer
2–4 weeks
Wireshark / Aircrack / KismetHackRF / SDRBurp SuiteOWASP IoT testing guide
33

RL & Adaptive
Control Systems

Reinforcement-learning controllers that beat hand-tuned PID, neural-network adaptive control, Model-Predictive Control with learned dynamics, Kalman filters fused with deep state-estimators. For systems where the plant is too nonlinear, time-varying, or unmeasured for classical control alone. Grounded in USF Control Systems coursework + concepts: feedback-control, state-space, stability, reinforcement-learning.

Robotics startups (manipulation / locomotion), drone manufacturers (high-agility flight), process-control engineers (nonlinear chem / pharma), motion-control integrators, autonomous-vehicle teams.

$10K–$30K build
+ $500–$1.5K/mo
2–6 weeks
MATLAB / Simulink ControlPython control libraryROS / ROS2PLC / OPC UA
34

Safety-Certified AI for Cyber-Physical Systems

AI systems that meet IEC 61508 / ISO 26262 / DO-178C — formal verification of LLM outputs, deterministic-latency inference paths, fail-safe state machines around model decisions, hazard / fault-tree analysis on agentic loops, certified deployment of ML in vehicles, medical devices, and industrial control. The intersection no consultancy is currently building because it requires both AI fluency AND classical safety engineering. Grounded in USF coursework explicitly named AI / Cyber-Physical Systems — taken as a single integrated course.

Medical-device manufacturers (FDA-graded AI), autonomous-vehicle / ADAS suppliers (ISO 26262 ASIL-D), aerospace (DO-178C DAL-A), industrial-safety integrators (IEC 61508 SIL-3). Anyone whose AI making a wrong call costs lives or licenses.

$25K–$80K build + $2K–$5K/mo
4–12 weeks
FreeRTOS / Zephyr / VxWorksTLA+ / SPIN / CoqMISRA C / SPARK AdaDO-178C / IEC 61508 processHazard / fault-tree analysis

Advanced AI Systems Architecture

35 – 40 / 40
EE Frameworks Productized — From a 559-Source Wiki of Coursework Synthesis
35

Multi-Agent Orchestration Architecture

Design and ship multi-agent AI systems using control theory — feedback loops with PID semantics (P=immediate retry, I=accumulated pattern learning, D=predictive halt), stability margins on agent-spawn cascades, circuit breakers as emergency stops, emergence detection as positive-feedback monitoring. For organizations whose 5+ agent stack goes off the rails, hits budget walls, or cascades into runaway spawning. Grounded in USF Control Systems + Electromechanical (33 sources) + my Rotating Bowl framework, governance-enforcer, and cognitive-monitor patterns running in production.

Companies with 3+ AI agents in production that no longer behave predictably. Anyone whose "AI workflow" became "AI casino."

$30K–$80K build
+ $2K–$5K/mo
4–10 weeks
PID feedback controlPole-zero stability analysisCircuit-breaker patternsFastAPI orchestration
36

AI Failure Engineering
(Radar Detection Theory)

Apply radar detection theory to AI failure monitoring. ROC-tuned alerting (Pfa vs Pd tradeoff — stop the alert fatigue), matched-filter recognition for known failure signatures, Kalman state estimation on runtime telemetry, Doppler-style detection of behavior-rate change (spend acceleration, error-rate creep). Real failure engineering, not Datadog dashboards. Grounded in USF Radar Systems (60 sources) + RF/Microwave Measurements (79 sources) + my emergence-detection-hook + failure-analyst + stuck-predicate framework.

Production AI teams drowning in alerts that don't matter and missing the alerts that do. SREs running LLM workloads.

$20K–$60K build
+ $1.5K–$4K/mo
3–8 weeks
CFAR detection thresholdsKalman filteringMatched-filter signaturesDatadog / Splunk integration
37

Agent Mesh Design (OSI for AI)

Design layered multi-agent architectures using network engineering. Routing tables (which agent gets what request), QoS classes (budget-pool prioritization), TTL (spawn-depth limits), congestion control (rate limits per agent tier). The "stack of agents" most teams ship is a mesh with no protocol — I bring OSI-model rigor. Grounded in USF Data Networks (13 sources) + Energy Delivery Systems (45 sources) + Wireless Mobile Comp Sec (15 sources) + my routing-monitor-hook + budget-pools + AGENT_TIERS production architecture.

Engineering orgs running 5+ AI agents that step on each other, retry endlessly, or drain budget unpredictably. Platform teams building agent infrastructure.

$25K–$70K build
+ $2K–$4K/mo
4–8 weeks
OSI-model agent layeringRouting tables (per-agent)QoS budget poolsFastAPI / async orchestration
38

Context Window DSP Engineering

Apply Shannon's channel-capacity theorem (C = B × log₂(1 + SNR)) to LLM context windows. The window IS a signal-processing channel — bandwidth (token limit), noise (irrelevant content), filtering (context surgery), windowing functions (Hamming/Hanning, not rectangular), Nyquist sampling theorem on token sampling rate. Typical result: 3–10× cost reduction with quality preservation. Public proof: sinc-LLM project itself, plus my token-nyquist + context-surgeon + nyquist_snr + nyquist_compress production tools. Grounded in USF DSP (65 sources) + Communications (75 sources) + Modulation (top concept, 121 source refs).

Companies with 5-figure-monthly LLM bills whose prompt cost grew 4× faster than usage. Anyone hitting context-window limits and assuming the answer is a bigger model.

$15K–$50K build
+ $1K–$3K/mo
2–6 weeks
Shannon channel-capacity mathNyquist samplingWindowing functionsSNR optimizationsinc-LLM 6-band format
39

Prompt Protocol Engineering

Design prompt protocols the way comms engineers design modulation schemes. Source coding (information-dense, structured formats per agent), modulation matched to receiver (SINC-2 prose for Opus, compressed keywords for Haiku), error correction (adversarial validation as repetition coding), forward-error-correction tradeoffs. Replaces "vibe-prompting" with engineered communication protocols. Grounded in USF Intro To Communication Systems (75 sources) + my SINC-2 format + mode-detect + adversarial-validate skills.

AI product teams whose outputs vary wildly between runs. Engineering teams that want a prompt spec, not prompt folklore.

$10K–$30K build
+ $500–$2K/mo
2–4 weeks
SINC-2 6-band formatSource / channel coding theoryAdversarial verification loopsPer-model modulation
40

Transformer Internals Audit & Custom Fine-Tuning

Read what the model is actually doing with your prompt — not what it claims it's doing. Attention-head analysis, embedding-space projections, Q/K/V matrix inspection, SVD-based LoRA design, eigenvalue audit of fine-tuning targets. When prompt engineering fails, this is the layer beneath it. For teams whose model "should work" but doesn't, where prompt-tuning has plateaued. Grounded in USF Engineering Analysis (34 sources) + linear-algebra concept depth (31 sources).

ML platform teams owning models in production. Companies considering fine-tuning vs. prompting and needing the right answer for their workload. Research teams whose models behave unexpectedly under load.

$20K–$60K build + $1K–$3K/mo
3–8 weeks
PyTorch hooks / TransformerLensLoRA / QLoRA / DPOSVD-based rank reductionEigenvalue / spectral analysisEmbedding-space probing

Languages & Markets

22 / 40
EN · PT · ES — Native Delivery
22

Multilingual AI Systems — EN / PT / ES

Most "multilingual" AI is English translated badly. Native delivery in English, Portuguese, and Spanish — voice agents, chatbots, content pipelines, document AI — built and tested by an engineer who speaks all three. Localized prompts, accent-correct voice, cultural register. Critical for Lusophone markets (Brazil, Portugal, Angola, Mozambique), Latin America, and US Hispanic businesses.

Brands expanding into PT/ES markets, government and enterprise contracts in Lusophone Africa, US businesses serving Hispanic customers, content teams localizing at scale.

$6K–$15K build + $400–$1K/mo
1–3 weeks per language pair
Native EN / PT / ESVoice cloning per localeCultural register tuningLusophone Africa expertise

Proof, Not Promises

// Real Production Systems

Public, verifiable engagements. Numbers from production logs, not pitch decks. More case studies (NDA-covered) available on the discovery call.

99%
16-phase content pipeline reliability

Production pipeline processing 500+ transcripts at 16 sequential phases (transcription, scoring, extraction, formatting, scheduling, publish). 99% completion rate at scale. Live at sr-demo-ai.com.

Content Extraction · Reliability Engineering
55hrs
Of "lost" coaching content recovered per month

Client had 55+ hours/month of unclipped Zoom coaching recordings. AI now identifies emotional peaks, framework drops, and contrarian takes — clips, formats, and queues them automatically. Content calendar fills itself.

Content Extraction Pipeline
12
Production tools exposed via custom MCP server

Self-hosted Model Context Protocol server (sincllm-mcp v2.0.0, SSE transport) giving Claude direct authenticated access to 12 internal tools and 3 resources. Architecture used in client deployments.

Custom MCP / AI Tool Servers
83K+
Lines of production Python shipped

Across SEO pipelines, browser automation infrastructure, payment-gated delivery, multi-tenant nginx deployments, and AI agent runtimes. Not POCs — running 24/7, processing real revenue.

Engineering Volume

Why Engineer-Built, Not Agency-Built

// The Honest Comparison

Three options in the AI services market right now. Here's where each one lands. I'll be the first to tell you when one of the others is the right call.

No-Code Agency Big AI Consultancy Hire In-House DLux Digital (Me)
Time to first ship 1–3 weeks 3–6 months 4 mo (hire) + 2 mo (ramp) 2 hours – 3 weeks
Total cost (year 1) $3K–$30K + monthly $200K–$2M+ $180K–$280K + benefits $2K–$40K build + $200–$5K/mo
Build approach Templated workflow stitching Slide deck → SOW → subcontract Greenfield from your hire Custom-built, BSEE-grade rigor
Who actually builds it Junior + n8n templates Subcontracted offshore team Your new hire (varies) Me. Senior, on every line.
Code ownership Locked to their workspace Their IP unless negotiated Yours Yours. Full handover, no lock-in.
Production reliability Demo-grade — fails at scale Variable Depends on hire Engineered for 99%+ reliability
When to choose One-off Zapier-tier task Fortune 500 transformation 10+ AI initiatives in flight Need it shipped, reliable, owned, fast

What Clients Say

// Documented Outcomes

These testimonials are representative examples based on documented project outcomes and typical client experiences. Names and company descriptions are illustrative — the engineering work, metrics, and methodology described are real.

Representative example

I had a backlog of long-form videos and no written assets. I expected the output to feel generic, so I held off. The system turned the backlog into 36+ publication-ready articles, and the 16-step workflow meant I wasn't chasing files or fixing formatting. It still sounded like my voice. If you want your content to exist beyond the video platform, this solved it.

Rachel Thornton

Creator & Host

Personal finance education channel

Austin, US

Representative example

We needed an automation pipeline we could ship into production, not a demo script. Mario delivered a 16-phase orchestrator with explicit JSON state, checkpoint recovery, and clean interfaces between phases. In our testing it handled failures predictably and resumed from the last good checkpoint instead of rerunning everything. Knowing he'd already put the pattern through 61 production runs gave us confidence, and our team integrated it into our stack with minimal rework.

David Chen

CTO

Developer tools startup

Vancouver, Canada

Representative example

Our content library had grown messy, and we didn't know what to consolidate. The authority audit surfaced a 42.55% consolidation upside and mapped which topics were actually carrying authority. The deliverable was clear, defensible, and easy to hand to writers and stakeholders.

Elena Varga

Director of Marketing

Mid-market software company

London, UK

Representative example

Lead qualification was the choke point in our client delivery. DLux built a 6-stage lead discovery and scoring pipeline that fit our intake process and produced 24 qualified leads we could action immediately. What mattered most was consistency: every lead came with the same fields, notes, and a score, so our team stopped debating what 'qualified' meant.

Sofia Morales

Agency Owner

Content and demand-gen agency

Mexico City, Mexico

Representative example

We brought Mario in to rebuild a brittle collector into something we could operate. He used Ray with async I/O, retries, and idempotent state so the job could recover cleanly. The design is explicitly sized for a 500K-item target and a 50 RPS tier, with backpressure and checkpoints instead of silent drops. He also handed over a MATLAB simulation suite (90+ files) that we used to sanity-check latency and fairness assumptions.

Arjun Nair

Senior Data Engineer

Payments and risk analytics company

Singapore

Representative example

I needed NLP capability I could resell without becoming an NLP researcher. Mario packaged a semantic analysis layer with a suite of 8 algorithms, plus clear inputs/outputs I could explain to clients. It let me deliver consistent audits without overpromising.

Tomas Novak

Independent Consultant

Compliance and policy advisory practice

Berlin, Germany

Try the Engineering

// Free Tools — Live Telemetry

8 free AI engineering tools that prove what production AI looks like — multi-model fallback chains, visible model selection, latency, SNR, and audit IDs. Same architectures I deploy for paying clients, sized for free-tier OpenRouter models. Three featured below — all 8 at sincllm.com/tools.

// 01 · Detection Theory

Hallucination Radar

Submit any LLM claim → 3 free models (Nemotron 120B + Gemma 31B + MiniMax) fact-check in parallel → see agreement matrix, ROC consensus, per-model citations of specific factual errors.

→ Try it free · Lead-in for service #36 ($20K–$60K)
// 02 · Control Theory

AI System Stability Auditor

Paste an agent prompt → returns pole-zero stability analysis, gain margins, positive-feedback loops that will runaway, and PID-style fixes (P=immediate, I=accumulated, D=predictive).

→ Try it free · Lead-in for service #35 ($30K–$80K)
// 03 · Functional Safety

AI Safety Hazard Analyzer

Describe an AI use case → returns IEC 61508 / ISO 26262 fault-tree, FMEA failure modes, SIL/ASIL recommendation, required safeguards, and a deploy / deploy-with-guards / do-not-deploy verdict.

→ Try it free · Lead-in for service #34 ($25K–$80K)
// 04 · Free PDF Checklist

10-Point AI Vendor Audit

A 10-criterion checklist to evaluate any AI agency before you sign. Yes/no questions covering monitoring, error budgets, ownership, fallbacks, drift detection, cost alarms, on-call coverage, data boundaries, hand-over. 15 minutes to run on any vendor.

→ Get it free · Lead-in for service #12 ($3K–$8K)
// 05 · Free PDF Framework

AI Build vs Buy Framework

A 10-criterion decision framework for engineering leaders. Each criterion resolves to a build path, a buy path, or a hybrid with a specific seam location. Includes scorecard worksheet, decision matrix, and quick-reference card. Run it in one architecture review.

→ Get it free · Lead-in for CTO advisory ($5K–$15K)
// 06 · Free PDF for CFOs

AI Cost Reality Check

A 9-question audit for CFOs whose AI line item doubled and nobody can name what doubled. Each question has a healthy / watch / bleeding state and the lever that recovers spend. Self-administered on a Friday morning. No API keys required.

→ Get it free · Lead-in for CFO Cost Sprint ($8K–$15K)
// 07 · Free PDF for CISOs

AI Incident Readiness Audit

A 12-control audit for CISOs and security leaders. Each control has a ready / gap / mitigation state. Names the failure mode, the install playbook, and the residual risk. The audit your auditor will cite next year.

→ Get it free · Lead-in for AI Safety Architecture Review ($10K–$20K)
→ See All 8 Free Tools

Strategy & Roadmapping

★ Premium Engagement
★

AI Strategy Consulting

Before building anything, most businesses need to know what to automate first and in what order. As a senior AI systems engineer, I map your highest-friction workflows, identify ROI, flag technical risks, and produce a prioritized automation roadmap — before a single line of code is written. Engineering depth that separates this from no-code agencies.

Any business that's heard the AI pitch a hundred times but doesn't know what's real, what's worth it, or where to start. Startup CTOs, ops directors, business owners.

$200–$500/hr
or $3K–$8K flat engagement
Automation roadmap + ROI model + vendor recommendations

How every
engagement
works

Five steps. Every service. Most systems are live within a week of first contact. Larger engineering builds run 2–6 weeks. Everything is documented and handed over — you own it. No platform lock-in, no vendor hostage situation.

Step 01
Discovery Call

30 minutes. We map the current workflow, identify the highest-friction points, and confirm the right service for the problem. No upsell — just clarity.

Step 02
Access & Audit

Read-only access to relevant systems. We audit integrations, data quality, and edge cases before touching anything. No surprises mid-build.

Step 03
Build & Configure

Custom build to match your brand voice, business rules, escalation logic, and existing tools. No templates — engineered from scratch.

Step 04
Test & Refine

Real-world scenario testing before go-live. You review every flow. Adjust until it's right — including edge cases and error handling.

Step 05
Go Live + Monitor

System goes live with 7-day monitored launch. Monthly retainer covers maintenance, model updates, performance tuning, and new feature additions.

Full Service Menu

All prices in USD. Retainer billed monthly. Build fee due 50% at kickoff, 50% at delivery. Strategy consulting billable hourly or flat-engagement.

#ServiceTierBuild FeeMonthly RetainerSetup Time
01AI Voice ReceptionistCore$3K – $5K$300 – $5002–4 hrs
02AI Review ManagerCore$500 – $1.5K$200 – $5001–2 hrs
03AI Lead QualifierCore$5K – $8K$500 – $1K4–8 hrs
04AI Ad Creative PipelineCore—$1K – $3K1 weekend
05AI Sales Call AnalyzerCore$2K – $4K$300 – $8004–8 hrs
06AI Document IntelligenceFlagship$8K – $20K$500 – $1.5K1–2 wks
07Predictive AnalyticsFlagship$10K – $30K$1K – $2K2–4 wks
08AI Compliance AutomationFlagship$15K – $40K$2K – $5K3–6 wks
09Edge AI DeploymentFlagship$8K – $25K$300 – $8002–5 wks
10Custom MCP / AI Tool ServersFlagship$4K – $10K$300 – $7001–2 wks
11Self-Hosted LLM StackFlagship$6K – $18K$400 – $1K1–3 wks
12AI System Audit & ReliabilityFlagship$3K – $8K flat—3–7 days
13Internal Knowledge AgentGrowth$5K – $12K$400 – $8003–5 days
14AI Proposal GeneratorGrowth$3K – $6K$200 – $4001–3 days
15Content Extraction PipelineGrowth$4K – $10K$500 – $1.5K1–2 wks
16SEO Article PipelineGrowth$5K – $15K$500 – $2K1–3 wks
17Payment & Access Control BridgeGrowth$4K – $10K$300 – $7001–2 wks
18Outbound Sales AutomationGrowth$5K – $12K$500 – $1.5K1–2 wks
19AI Avatar & Video PipelineGrowth$3K – $8K$400 – $1K3–7 days
20Custom Browser AutomationGrowth$4K – $10K$400 – $1K1–2 wks
21Telegram Bot InfrastructureGrowth$2K – $6K$200 – $6003–7 days
22Multilingual AI Systems (EN/PT/ES)Markets$6K – $15K$400 – $1K1–3 wks
23AI-Driven Smart Building EnergyAI×EE$12K – $35K$800 – $2K3–6 wks
24AI-Driven Industrial Predictive MaintenanceAI×EE$15K – $40K$1.5K – $3K4–8 wks
25AI-Assisted Solar PV & Microgrid DesignAI×EE$8K – $20K$400 – $8002–4 wks
26Embedded AI & TinyML FirmwareAI×EE$15K – $45K$500 – $1K3–8 wks
27AI × DSP — Signal IntelligenceAI×EE$5K – $15K flat$200 – $400/hr1–4 wks
28AI-Powered Engineering Document AutomationAI×EE$10K – $25K$500 – $1.2K2–5 wks
29ML-Augmented RF & SDR SystemsAI×EE$15K – $50K$800 – $2K3–8 wks
30AI-Optimized Grid & Power ElectronicsAI×EE$15K – $50K$1K – $2.5K3–8 wks
31AI-Enhanced Radar & DetectionAI×EE$20K – $60K$1K – $3K4–10 wks
32AI-Powered Wireless & IoT SecurityAI×EE$8K – $25K$2K – $5K2–4 wks
33RL & Adaptive Control SystemsAI×EE$10K – $30K$500 – $1.5K2–6 wks
34Safety-Certified AI for CPSAI×EE$25K – $80K$2K – $5K4–12 wks
35Multi-Agent Orchestration ArchitectureAI Arch$30K – $80K$2K – $5K4–10 wks
36AI Failure Engineering (Radar Detection)AI Arch$20K – $60K$1.5K – $4K3–8 wks
37Agent Mesh Design (OSI for AI)AI Arch$25K – $70K$2K – $4K4–8 wks
38Context Window DSP EngineeringAI Arch$15K – $50K$1K – $3K2–6 wks
39Prompt Protocol EngineeringAI Arch$10K – $30K$500 – $2K2–4 wks
40Transformer Internals Audit & Fine-TuningAI Arch$20K – $60K$1K – $3K3–8 wks
★AI Strategy ConsultingPremium$3K – $8K flat$200 – $500/hr1 week

Frequently Asked

// Honest Answers
How fast can you actually ship?
Core services (AI Voice Receptionist, Review Manager, Lead Qualifier, etc.) ship in 1–8 hours. Engineering-grade builds (Document Intelligence, Compliance Automation, Edge AI) take 1–6 weeks depending on integration complexity. Most clients see a working demo within 7 days of the discovery call.
What if I don't know which service I need?
Start with AI Strategy Consulting ($3K–$8K flat). I map your highest-friction workflows, identify ROI, flag technical risks, and produce a prioritized roadmap before a single line of code is written. Often the audit alone reveals that you don't need what you thought you needed.
Do you handle the AI/LLM costs, or do I?
You own the API keys and pay providers directly (Claude, OpenAI, ElevenLabs). I never mark up tokens — that's vendor lock-in dressed up as a service. For high-volume clients, I'll often recommend Self-Hosted LLM Stack (#11) so per-token costs go to zero after deployment.
What does the monthly retainer actually cover?
Monitoring, model updates (when better models drop), prompt tuning, error-handling improvements, edge-case fixes, and small feature additions. It's an SLA, not a subscription. If a system is rock-solid and you need nothing, you can pause the retainer — and resume when you do.
Can I see the code? Do I own it?
Yes to both. Every build ships with full source code, deployment scripts, and runbooks. You own everything. No platform lock-in, no proprietary "agency dashboards" you can't access without my login. If we part ways tomorrow, you keep operating.
What's the difference between you and a no-code agency using n8n?
No-code agencies stitch templates. I architect systems. When their workflow breaks at 10x volume — which it will — they can't fix it because they didn't build the underlying logic. I write production Python with monitoring, error budgets, and explicit success metrics. See the comparison table above.
Do you work with non-US businesses?
Yes — globally. I deliver natively in English, Portuguese, and Spanish (see service #22). Strong specialty in Lusophone markets (Brazil, Portugal, Angola, Mozambique) and Latin America. Time zones overlap from US Pacific through Western European business hours.
What if my project doesn't fit any of the 40 services?
Mention it on the discovery call. The 40 services are the most common asks — but I've shipped custom systems for outliers too. If it's an engineering problem with an AI/automation answer, it's probably in scope. If it's not, I'll tell you and recommend who to talk to.
How do payments work?
50% at kickoff (covers discovery, audit, architecture). 50% at delivery (after you've reviewed and approved). Stripe, ACH, or wire — your call. Monthly retainers billed automatically. No prepay required for retainers — first month is invoiced after go-live.
What does "BSEE-grade rigor" actually mean?
Electrical engineering on physical systems means a mistake makes the building go dark. You don't get to ship a "demo." You design for redundancy, fault tolerance, and predictable failure modes — and you prove it works under load before energizing. I bring that same discipline to AI: monitoring on every critical path, explicit success metrics, error budgets, fallback paths, drift detection. Most "AI agencies" ship demos. I engineer for production.

Recent Thinking

// From the Field

Working notes from production AI engineering — what I learned shipping real systems for real businesses. Full archive on LinkedIn.

2026 · LLM Communication

The LLM is the Genie from Aladdin

Make one wish — and you get exactly what you asked for. But all wishes need to be in structured, properly formatted prompts. I was always leaving out important points because I felt like the model would read between the lines. I was wrong. Then I asked the model to change a single line of code and it spent 80k tokens. That's when I realized: tell the genie exactly where you want the change, with a strong format prompt.

Read on LinkedIn →
2026 · AI Hallucination

If You Don't Understand It, You Are the Second Hallucination

If you don't understand what you're asking the AI to do, then you become a second hallucination source — and now there are two working on your project. Your complete understanding of what the AI is doing is the only gate between hallucination and reality. This is why every system I ship has explicit success metrics: they're the gate.

Read on LinkedIn →
2026 · Content Operations

Your Best Content Is Rotting on a Hard Drive

One of my clients had 55 hours of Zoom recordings sitting unclipped. That's four months of daily content, fully produced, completely invisible. The bottleneck was never creation. It was extraction — turning raw recordings into publishable pieces. The gold is already mined. It's just sitting in a pile, unsorted. That's the system I build.

Read on LinkedIn →
// Let's Build Something That Works

Tell me what's breaking.

Discovery calls are 30 minutes, free, no pitch. We'll map your current workflow, find the highest-friction point, and confirm whether one of the 40 services fits — or whether you need something custom. If we're not the right fit, I'll tell you who is.

Book a Call LinkedIn Email Direct
— or send the brief —

// Limited capacity. Currently accepting 4–6 new engagements per quarter.