MatheusMello
I build AI systems that run in production and keep working after launch. That means agent architectures with controlled hand-offs, retrieval that stays inside the caller's data boundary, evaluation harnesses wired into CI, and cost telemetry that makes spend legible.
I also ship my own products. Doozle is my app and game studio; SoleOS is an analytics dashboard I built and operate. Running my own portfolio is why I care about the unglamorous parts — regression evals, failure modes, latency budgets, the cost of a model change — rather than the demo.
Engagements are fixed-price against written acceptance criteria. You know the scope and the number before anything starts.
Five problems.
Not every AI discipline — these five. They are the ones I hit repeatedly running production systems and my own products.
- 01
Production agent reliability
What actually breaks once an agent leaves the demo: tool-call failures, silent regressions, and the recovery paths that keep a run alive.
- 02
Multi-tenant RAG
Enforcing per-tenant scope through the data and retrieval layers, so one tenant's query cannot reach another tenant's records.
- 03
LLM evaluations in CI
Fixture eval sets and regression evals that run on every change, so a prompt or model swap cannot quietly degrade output.
- 04
Cost-aware model routing
Token budgets, caching, and multi-tier routing — treating inference spend as an engineering constraint rather than a surprise.
- 05
Operating a portfolio of AI products
What it takes for one operator to build, ship and run several AI products at once, and which systems make that possible.
I ship my own.
Consulting on production AI while never operating anything is easy to say and hard to back. These are mine — built, shipped and maintained.
- App & game studio
Doozle
- My studio for consumer applications and games, designed, built and shipped to the app stores.
- Analytics + AI brief layer
SoleOS
- A multi-project analytics dashboard that consolidates metrics across projects, with an AI layer that turns the raw numbers into a written read of what changed.
- iOS & Android
Consumer apps
- A portfolio of shipped mobile apps across nutrition, habit-tracking and fitness — the products that keep me honest about what production actually costs.