Production AI engineering · sinc-LLM

AI systems built for the work your business depends on.

I design, build, and audit production AI systems—from workflow automation to private AI infrastructure. Start with one workflow, a defined scope, and a system you own.

Book a 30-minute call Check a workflow free

Mario AlexandreAI Systems EngineerAbout Mario

Your workflow
Inputs and tools
Review and controls
Observable output

Illustrative workflow. The call defines the scope for your project.

Start with the work that needs to get done.

Choose a build, an audit, or a private AI system around one defined job.

Automate a repeated workflow

Outreach and browser workflows, built around your existing tools.

Explore workflow builds

Review an existing AI system

Architecture, vendor, security, and cost work with defined deliverables.

Compare audits

Build private AI infrastructure

Local knowledge systems, observability, evaluation, and action gates.

Explore infrastructure

Defined scope. Visible price.

Compare the current fixed-price builds and audits before choosing the next step.

Compare fixed-price services

From one workflow to a documented handover.

The engagement starts by clarifying the job, its inputs, and the work to be delivered.

Scope one workflow
Review data and access
Build and test
Document and hand over

Inspect the work in the open.

The original public repository links below let you inspect the code and project documentation.

Read the public code

Bring one workflow to the conversation.

Use the existing call to discuss fit, scope, and the next engineering step.

Book a 30-minute call Check a workflow free

Built onvLLMMCPClaude APIOpenAIWhisperElevenLabs

Built in the open.

// Open source on GitHub, all repos ↗

// Lean subagents

Lean agents for Claude Code

Two drop-in subagents: a worker and a blind checker. On my setup, they start at about 7.5k tokens versus 131k for general-purpose.

GitHub ↗★ 14

// Multi-agent orchestration

crews

Plans Claude Code subagents by model and effort, and guards every dispatch.

GitHub ↗★ 4

// Cross-agent allocation

olondunge

Hands work across Claude Code, Codex and Grok, then has a different model verify it blind.

GitHub ↗Free

// Prompt engineering

sinc-llm

Nyquist-Shannon sampling applied to LLM prompts: six specification bands.

GitHub ↗★ 9

// Agent guardrails

python-evidence-gate

Stops an AI agent from calling Python work done without passing ruff, mypy and pytest.

GitHub ↗Free

// Sampling policy

multi-shot-intelligence

Decides whether to sample a model N times, and how. Checked against MATLAB.

GitHub ↗Free

// Long-running agents

journey-map-thinking

Position tracking for long-horizon agents. Ships with its own null result.

GitHub ↗Free

// Prompt compiler

question-engine

Turns a raw prompt into a fully specified execution prompt.

GitHub ↗Free

// Agent voice

codex-soul

A voice contract for Codex: human, precise, trustworthy prose.

GitHub ↗★ 1

This is not an AI capability problem. It is a production readiness problem.

// The honest read

Years EE foundation

83K+

Lines production Python

// No-code agency

Fast and cost-effective

Good at workflow automation. Optimizes delivery speed, not error budgets or fallback paths.

// Vendor implementation team

Configures it in days

Their incentive is adoption. Handover happens on their timeline, not yours.

// Build internally

Silent failures

No exception, no stack trace: specification aliasing. First-party research ↗

You own the code. Full handover, no platform lock-in.

Selected AI engineering work.

// Selected work

// SEO engineering

SEO

Technical SEO tooling

Custom SEO engine for indexing, schema, and internal linking.

// Content automation

Pipeline

Article research and distribution

Research, writing, and distribution in one pipeline.

// Crawl & indexing

Indexing

Search discovery infrastructure

Websites

Site engineering

AI engineering websites with schema, sitemap, and IndexNow.

// Outreach automation

Messaging

AI SMS outreach

Production SMS engine on a $50/month plan.

// Cost engineering

Routing

Local and cloud model selection

Local models for selected work, with cloud routing for other tasks. Compare per-request cost, output quality, and data boundaries for the workload.

// Browser automation

Browser

Automation for web applications

Browser automation using real clicks, typing, and persistent sessions across web applications.

// RAG & content infra

Retrieval

Machine-readable site knowledge

RAG grounding plus a machine-readable llms.txt layer.

// Private knowledge engine

Knowledge

Records and QA workflow

Self-hosted model and a QA agent in the data-processing workflow.

Proof you can inspect.

Source-inspected tools, recorded workflow activity, and research with a public source.

Engineering toolkit

155+

source-defined MCP tools

Across 17 inspected server implementations, spanning orchestration, research, engineering checks and content workflows.

Verified minimum from source. Client deployments are not counted here.

Source inspected · 10 Oct 2026

Explore the tool inventory

Counted 155 explicit named tool definitions across the 17 implementations below. Each server is counted once; tools are counted within their server.

Archived versions, test files, copied repositories and dynamically registered tools are excluded. This is a source inventory, not a runtime availability or production reliability claim.

Maestro17 tools
Engineering Workbench18 tools
Simulation10 tools
Codex Crews15 tools
Knowledge18 tools
Execution12 tools
Airbnb Rank1 tool
PriceLabs Sync1 tool
Grok5 tools
HKOES11 tools
ESOS Gap Map4 tools
ESOS Grounding1 tool
D2K6 tools
Allocation7 tools
UI Stack18 tools
Jev4 tools
Blaster Article7 tools

Recorded workflow activity

1,141

Blaster jobs marked done

Distinct job IDs in the retained local logs of the research-to-article workflow.

Includes development, retry and repair activity.

Log snapshot · 8 Oct 2026

Read the measurement scope

Of 1,687 distinct job IDs in the selected retained queue snapshots, 1,141 have only the status “done”. IDs are deduplicated across retained database copies.

Queue submissions span April to September 2026; submission timestamps are not completion times. The logs include development, retry and repair activity.

This measures recorded local activity. It does not establish an overall success rate, unique transcripts, published articles, customers or time saved. Private client content is not exposed.

Published research

460

total paper downloads

Sinc Reconstruction for Language Model Prompts: the published specification-axis research.

373 unique downloads · Zenodo record statistics.

Verified snapshot · 10 Oct 2026

Check the public source

The Zenodo record reports 460 total downloads and 373 unique downloads in the saved snapshot. These are record statistics, not a count of clients, installations or unique people.

Download totals change over time. The complete paper title and DOI appear below.

View Zenodo statistics

Research you can read.

Explore the methods behind the work: agent navigation and prompt specification.

Prompt specificationResearch paper

Sinc Reconstruction for Language Model Prompts: Applying the Nyquist-Shannon Sampling Theorem to the Specification Axis

Nyquist-Shannon sampling applied to the prompt specification axis.

460

Total downloads373 unique downloadsZenodo · verified 10 Oct 2026

Agent navigationResearch paper

Journey Map Thinking: A Persistent Navigation Layer for Long-Running AI Agents

A persistent navigation layer for long-running AI agents.

—

DownloadsActive record pending

Journey Map Thinking: the original Zenodo record was removed as a duplicate. Download statistics await the active record.

Download counts are a dated snapshot of Zenodo statistics.

The services, methods, and examples described here do not establish an expected performance result for your workload.

Start a conversation

Let’s put your
workflow to work.

Bring one workflow to a 30-minute call. We’ll discuss the problem, scope, and next engineering step.

Book a 30-minute call

Explore first? View services & pricing

Call directly(813) 820-8270 Email directlymdalexandredlux@gmail.com

Connect on LinkedIn

Prefer to write?

AI systems built for the work your business depends on.

Start with the work that needs to get done.

Automate a repeated workflow

Review an existing AI system

Build private AI infrastructure

Defined scope. Visible price.

From one workflow to a documented handover.

Inspect the work in the open.

Bring one workflow to the conversation.

Built in the open.

Lean agents for Claude Code

crews

olondunge

sinc-llm

python-evidence-gate

multi-shot-intelligence

journey-map-thinking

question-engine

codex-soul

This is not an AI capability problem. It is a production readiness problem.

Fast and cost-effective

Configures it in days

Silent failures

Selected AI engineering work.

Technical SEO tooling

Article research and distribution

Search discovery infrastructure

Site engineering

AI SMS outreach

Local and cloud model selection

Automation for web applications

Machine-readable site knowledge

Records and QA workflow

Proof you can inspect.

source-defined MCP tools

Blaster jobs marked done

total paper downloads

Research you can read.

Sinc Reconstruction for Language Model Prompts: Applying the Nyquist-Shannon Sampling Theorem to the Specification Axis

Journey Map Thinking: A Persistent Navigation Layer for Long-Running AI Agents

Let’s put yourworkflow to work.

Let’s put your
workflow to work.