Real-time critical systems
C++/Qt, SIP/VoIP, WebRTC, and Asterisk across emergency and call center platforms — SICO, SAU-Web, and the CNPq CRM. High capacity, low latency, and uptime as a requirement rather than a target.
Critical systems · AI in production
I take AI to production — and build the governance it demands.
I'm Wendel Maques, a software engineer. I spent 12+ years building real-time systems that cannot go down: emergency telephony, high-capacity call centers, and ambulance dispatch. Today I apply the same discipline to AI — speech and language pipelines over real data, GPU inference, and the decision trail that audit, legal, and the board actually ask for.
- Mission-critical software: ~10k calls/day at 99.99% uptime.
- AI on real data: ~192 GB of audio processed through a pipeline I built.
- Governance that survives an audit — not a loose spreadsheet.
What I help with today: cutting AI regulatory exposure and unblocking executive decisions, without slowing the operation down.
What I do
Three fronts that hold each other up: the system has to work, the model has to be right, and the decision has to be defensible.
AI in production
ASR and NLU pipelines in Brazilian Portuguese: Whisper Large v3 fine-tuned for medical emergency, BERTimbau for risk triage and symptom extraction, and a vLLM inference API across multiple GPUs.
Platform and operations
Cloudflare Workers, AWS, MySQL with replication, CI/CD, and versioned infrastructure. Products of my own running in production — from deploy to being on call when it breaks.
AI governance consulting
For companies already running AI in production. Fixed scope, defined deliverable, timeline and price on the table before we start.
Compliance assessment
We map flows, data, decisions, and legal risk in 7 days. You get a criticality matrix and an initial 30-day plan.
BRL 4,900Controls implementation
Approval controls, usage policy, prompt auditing, and the minimum trail for day-to-day operational governance.
Scope-basedTeam training
Hands-on training for product, legal, and engineering teams, with routines, checklists, and simulations on real cases.
From BRL 6,500How I work
A short method that avoids rework and builds internal trust with real evidence.
1) Discovery and mapping Interviews, an inventory of the AI assets in use, and data flows — to find the exposure points fast.
2) Executive prioritization Ranking by legal, operational, and financial impact — what goes first and what can wait.
3) Controls and delivery Implementation, initial monitoring, and an executive report ready for a go/no-go decision.
Evidence and artifacts
What you walk away with — material that works in an audit and in an executive review.
Why this becomes urgent now
This is not about owning a pretty tool. It is about cutting concrete risk with evidence that holds up for audit, leadership, and operations.
LGPD baseline for protecting and reviewing data used in AI models
AI Act getting ahead of risk criteria and traceability requirements
Compliance clear communication between legal, product, and engineering
Let's talk
Fill in the form. Within 48 business hours I reply with an initial analysis and the next steps.
What you get A risk matrix, priorities, and a 30-day plan with named deliverables.
Who it is for Teams with AI in production that need a decision trail and audit-ready evidence.
No noise The goal is executive clarity: what to do, why, and in what order.
Privacy: I use your data only to answer this message. No automated mailing, no sharing with third parties.