Senior Product Engineer
AI-native product engineer with 12 years of experience shipping end-to-end solutions.
Who is Andre Treib?
I’m a staff software engineer from Brazil with 12 years of experience. I started in 2014 as a junior engineer at a vehicle-monitoring company, and along the way I took on the product side too: I introduced agile practices as a Scrum Master, owned a roadmap as a Product Owner, and stepped in as the PM on projects when my team had no capacity for one. I ask what a feature is for before I ask how to build it.
For the last five years I have worked remotely on English-speaking teams, mostly at early-stage startups, where I usually own a system from the first design doc to the production rollout. Today that means applied AI, with one rule I keep: the model is never the source of a value. Code holds the facts, and every decision comes from a measurement, also when the answer is no.
I like to teach. I mentor engineers on how to test, and I point them to the code and the docs instead of handing them the answer. I’m also doing a post-graduation in AI engineering.
Hiring for an early-stage team or an AI product?
Email me. A few lines about the team and the problem are enough to start.
Prefer LinkedIn? Message me on LinkedIn.
Remote from Brazil (UTC−3), which overlaps US working hours. Five years on English-speaking remote teams. Native Portuguese, fluent English, intermediate Spanish.
Where Andre Treib has been.
12 years of full-stack delivery — last 5 on English-first remote teams.
House Numbers
CurrentSenior → Staff Software Engineer
I was the first engineer hired at House Numbers, a US mortgage-software startup, and I’m now the de facto tech lead of its AI platform. Lending is regulated and a wrong answer is costly, so I build LLM systems where code holds the facts, the model reasons over them, and every step is measured.
- Shipped
A real-time voice AI agent that loan processors talk to like a colleague
Processors ask about a loan out loud and get a spoken answer, with first audio in about 0.9 s. I built and shipped the agent. Human-in-the-loop guardrails live in code: any change it wants to make to a loan becomes a proposal that a person approves.
How it works (Real-time voice AI agent)
- Each stage streams into the next. Streaming ASR transcribes while the processor talks, and semantic endpointing reads that live transcript to decide when a sentence is finished. A clear question gets a fast reply, and a trailing “and…” gets more time.
- Two-stage barge-in lets a person interrupt. The agent’s voice ducks at once and stops only after sustained speech, which fixed a user complaint that typing cut the agent off.
- Speculative RAG starts the vector-database search on the partial transcript. Events from the platform’s event-driven architecture keep that index current.
- The LLM calls tools over MCP, and streaming TTS speaks the reply as it arrives.
Automated tests cover the whole pipeline, including LLM evals that catch regressions and A/B tests that decide which changes ship.
- In productionCo-designed
An AI agent for loan processing that saves 12+ hours of manual work per loan
I co-designed the core AI agent: 17 event-driven workflows in production that mix deterministic code with LLM steps. Guardrails gate every action. Only high-confidence, low-risk actions run on their own, and the rest wait for a loan processor’s approval.
How it works (Core AI agent for loan processing)
- Workflows react to loan events or run on a schedule.
- Each workflow calls a custom MCP server with idempotent tools. An event-driven system can deliver an event twice, so a repeated call must never apply the same change twice.
- I researched, tested and introduced the “reason-then-execute” pattern, now the team default. The agent first writes down what should happen and why, and only then calls tools against that plan, so each run can be audited.
- Evals track accuracy.
- R&D · not in production
Deterministic lending rules from a 1,196-page guide
Fannie Mae’s Selling Guide sets which loans it will buy and on what terms. I led R&D on a pipeline that turns the guide and its pricing and eligibility matrices into deterministic rules, so a query returns the one table cell that fits a loan instead of handing an LLM the whole table. I measured every iteration to hold zero false positives.
How it works (Lending-guide rules pipeline)
- Each table is read from the PDF’s layout, validated and stored as typed conditions and values.
- The model only interprets structure, since its output varies between runs. Code copies every value.
- Measuring every run that way also let me remove an LLM-review stage that added cost and hurt stability.
How I work
I work closest to the CEO and PM. I take on the loosely defined projects, scope them, and coordinate and mentor the other senior engineers who build them. I step in on production incidents such as OOMs, memory leaks and cache stampedes. Underneath the AI work, I build event-driven microservices in TypeScript, Node.js and MongoDB, tested at unit, integration and E2E level. My own coding agents run through a review workflow I built, which has completed 326 tasks across 8 projects since September 2026.
StackTypeScript · Node.js · MongoDB · Qdrant · MCP · Vercel AI SDK · Docker · Kubernetes · AWS
Avenue Code
Senior Software Engineer
Embedded with a San Francisco client to design and ship a Next.js + serverless SaaS end-to-end (TypeScript, PostgreSQL, MongoDB, Auth0, Lambda) — stood up the CI/CD pipeline on GitHub Actions, onboarded 3 engineers, and delivered my first professional Rust engagement.
Eldorado Research Institute
Mid → Senior Software Engineer
Full-stack engineer on a visual-computing and ML product — built the React/Node/Kotlin interface against ML inference modules with TDD and Clean Architecture end-to-end, then took over as Product Owner, owning roadmap, quality and delivery cadence.
MachadoTI
Mid Software Engineer & Account Manager
Technical reference helping an IT-infrastructure business break into software delivery — shipped outsourced projects across C#, PHP, React, Node, Ionic and SQL/NoSQL for 10+ clients while owning team, cost and deadline management plus client communication.
Digicounter
Junior → Mid Software Engineer & Scrum Master
Drove the full refactor and redesign of the core vehicle-monitoring SaaS, shipped a mobile app — including the ETA algorithm — serving 200+ vehicles and 1,000+ daily passengers, and brought agile methodology into the company.
Shipping AI products in public.
A selection of products I've shipped — the AI work I keep sharpening on, plus the side projects that came before it.
Waypoint
Anonymous travel reviews on a world map. Search any city, filter ten place categories, and read LLM-summarized “what travelers say” digests — a deep-linkable MapLibre map on Next.js.
Maragato Labs
My AI-native software studio — tradition and technology from the pampas. Conversational systems, knowledge agents, and autonomous workflows, presented in a trilingual React Router v7 site with generative pixel-art.
Tripwise
Next.js 15 trip planner powered by an LLM. Plan a trip end-to-end — destinations, itinerary, packing — through a single conversational flow.
Jerry
AI-powered 3D character built with Next.js, Three.js and OpenAI. Open-source experiment exploring real-time voice interaction with a rigged 3D model.
E-book Factory
One-shot ebook generator on top of an LLM with an image-generation pipeline. Type a topic, get a fully-illustrated ebook back.
Nutri SaaS
End-to-end SaaS for nutritionists with patient management, smart calendar, meal planning, anthropometric assessment, and an embedded per-patient AI assistant.
Easypack
Travel packing assistant that takes the trip details and returns the packing list — never forget what to bring again.
Bairro Aberto
Glassdoor for neighborhoods. Reviews, ratings, and lived-in signal for places before you move there.
Cloudret
Open-source visual cloud-infrastructure builder. Swap between diagrams and a human-friendly DSL, then export Terraform — fully client-side so user data never leaves the browser.
The Background Text
Open-source browser-side image editor. Uses a WASM model to extract foreground, then composes text behind the subject. Runs entirely client-side.
What people say about Andre Treib?
Check out some testimonials from people I have worked with.
André has been my closest partner and my go-to on nearly everything since day one. As the first engineer hired to build our platform, he knows it more deeply than anyone, and he's also deeply product-minded, so much so that I never felt like the only PM in the room. He pushes us to think outside the box and try new things, while staying grounded in the requirements and what the team actually needs. He goes above and beyond, nights, weekends, and overtime, to solve the hard problems and make sure the team succeeds, and he makes the impossible look easy. Every decision he makes is thoughtful and backed by real data and research. Mortgage is a notoriously complex space to build in, dense with domain knowledge, legal, and compliance requirements, and André has never shied away from it. He meets it with curiosity, always eager to learn and grow his skill set, and he's constantly testing new tooling and AI models to work smarter.
He's bridged roles more than once, stepping in to PM projects himself when we didn't have the capacity, and mentoring more junior engineers by sharing his time, his deep experience, and his knowledge of our tooling, systems, and process. He's made the product better, our outcomes better, and the whole team better. I can't imagine what this stretch would have been like without him, and he's someone I'll actively look to work with again and again throughout my career.
Crystal ButmanProduct Manager at House Numbers
Andre brings a positive can-do attitude to his work, which is refreshing and motivating to be around. In terms of technical capabilities, he is very equipped to handle the work and is a continued advocate to improving the code base and the patterns used within it. He understands that it is appropriate to invest in best practices, but also at times it is best to drive toward near term goals and backfill missing pieces. Andre also appears to be a natural people organizer, getting everyone aligned toward success.
Brian GlynnIAM Pulse
Andre is one of the most talented and qualified professionals I have worked with so far. He does his work extremely well, with dedication and pro activity, and always with a great mood. We briefly worked together at Eldorado Research Institute and he helped me on my first steps there. Someone with high potential to achieve top ranks in any company due his commitment, respect, openness, focus and courage. It's impossible to not recommend his work. He is totally above average.
Mario KollerEldorado Research Institute
Andre did a great job at Eldorado Research Institute where I had the chance to work with him. He's an excellent software developer and delivered all requests with a lot of quality and timely. He also acted as a real facilitator inside the team with his positive attitude and proactive behavior. I strongly recommend Andre for any company looking for an outstanding software developer!
Thiago SilvaEldorado Research Institute