Applied AI & Machine Learning
Research, modeling and evaluation, from the paper to the inference server. Generative models, decision systems and ML under real constraints.
Independent research & engineering lab
Systems where failure is expensive
Belabs works at the intersection of machine learning, low-level systems in Rust and financial infrastructure. We build software that has to be correct, fast and secure, and we keep quiet about it.
- Disciplines
- ML · Rust · Fintech · Crypto
- Disclosure
- NDA by default
- Base
- Brazil · Remote
- Status
- Accepting projects
01 / Practice
Four disciplines, one concern: getting it right.
The work spans research and production. The common thread is that mistakes in these systems cost money, data or trust.
Low-level Systems in Rust
Services and libraries where latency, memory safety and predictability come first. Profiling, concurrency and careful use of unsafe.
Payments & Financial Infrastructure
Ledgers, settlement, reconciliation and the integrations around them. Designed for audits, failure and regulators.
Cryptography & Distributed Systems
Blockchain integration, key management, signing flows and consensus, treated as engineering rather than hype.
02 / Method
Operating principles.
Nothing we build can be shown here, so this is how we build it.
- Threat model before code.Every system starts with who could break it and how. The design follows from that.
- Correctness is a feature.Types, invariants, property-based tests and, when it pays off, formal reasoning.
- Few dependencies, all understood.Every crate, package and service is part of the attack surface and is reviewed as such.
- Measure, then optimize.Benchmarks and profiles decide performance work. Opinions don't.
- Write it down.Decisions, trade-offs and failure modes are documented so the system outlives its authors.
03 / Work
Selected work.
Most of our work runs inside systems that move money or protect sensitive data. We don't disclose clients, architectures or code. These are the kinds of problems we take on.
Payment & settlement infrastructure
Ledgers, reconciliation and settlement pipelines in regulated environments.
ML systems in production
Models researched, evaluated and deployed under real latency and cost constraints.
Security-critical Rust components
Low-level services where memory safety and auditability are not optional.
Cryptographic & blockchain systems
Key custody, signing flows and on-chain / off-chain integration.
→ References are available privately, under NDA.
04 / Notes
Technical memoranda.
Open research and essays. The concepts are shareable even when the systems aren't.
- BL-TM-001Artificial Intelligence: a realistic look at its past, present and futureEssay on what AI is, what it isn't, and where it is going.2024
- BL-TM-002FLOW models: foundations, applications and advances in image generationSurvey of flow-based generative models.2024
- BL-TM-003MaskDecide: diffusion models for structured decisionsOpen-source research project.Open source
05 / About
A small lab, on purpose.
Belabs is an independent research and engineering lab founded by Fabio Covolo Mazzo, a researcher in machine learning and AI and a low-level Rust developer. His career spans data science, payments and blockchain.
The lab stays small so that every engagement is handled directly, with the care you'd want for a system your own family depends on.
06 / Contact
Have a hard problem?
contact@belabs.cloud
Tell us what you're building and what can't go wrong. Encrypted communication is available on request.