Lead developer / ML & AI engineer
FinRePAI · BitGov
AI credit-risk evaluation for public-sector suppliers
An AI platform that evaluates the financial risk of Brazilian public-sector suppliers. Continuation of FINAI-BitGov, developed at IFCE's Innovation Hub in partnership with Bitgov — multi-agent data extraction, supplier rating and explainable analytics.
FinRePAI · BitGov is an AI platform that assesses the financial risk of public-sector suppliers in Brazil, developed at IFCE's Innovation Hub with EMBRAPII support in partnership with Bitgov. Continuing the earlier FINAI-BitGov work, it combines classical machine learning with a multi-agent LLM system. Adaptive agents extract data from public transparency portals — 200+ sources today, with an architecture designed to scale toward all 5,500 Brazilian municipalities. On top of that data sit several purpose-built modules: SiRaC, a four-dimension supplier rating system using exponential smoothing and survival models; MoBaVE, for backtesting and empirical validation; and RExAn, which generates explainable analytical reports via LLMs. The platform uses fuzzy matching to link payments to contracts and ML to predict manager evaluations, giving administrators an auditable, explainable view of supplier risk on a Python 3.12, Django and XGBoost stack.
Problem
Public bodies must evaluate supplier financial risk across thousands of municipalities, but the data is scattered across transparency portals and the reasoning has to be explainable and auditable.
Approach
Adaptive multi-agent extraction feeding ML and survival models — SiRaC (rating), MoBaVE (validation) and RExAn (LLM-generated explainable reports) — built on Python 3.12, Django and XGBoost.
Outcome
An active platform extracting from 200+ public sources and architected to scale to 5,500 municipalities, developed under IFCE Innovation Hub · EMBRAPII · Bitgov.
Highlights
- Adaptive multi-agent extraction from public transparency portals (200+ sources, scalable to 5,500 municipalities)
- SiRaC — supplier rating system (4 dimensions, exponential smoothing, survival models)
- MoBaVE — backtesting and empirical validation module
- RExAn — explainable analytical reports generated via LLMs
- Fuzzy matching to link payments to contracts; ML prediction of manager evaluations
Metrics
- Institutional context
- IFCE Innovation Hub · EMBRAPII · Bitgov
- Core package stars
- ★3 (bitgovprocessor)
Stack
- Python 3.12
- Django
- XGBoost / ML
- Multi-agent LLM system
- Survival models
- RESTful APIs
- uv