Applied AI · Finance · Open Source

Production-grade AI for fintech data teams:
clone it, run it, ship it.

Open source systems solving the most expensive data problems in financial services: regulatory compliance, customer risk, credit decisions and outhers. Each one ships with architecture, working code, and a path to production.

Browse the Projects →
Live in the Laboratory
[GENAI-001] BSA/AML Regulatory RAG
Gemini · ChromaDB
[RISK-001] Customer Churn Scoring
XGBoost · SHAP
[CAUSAL-001] Credit Limit Intervention
DoWhy · EconML
3 of 3 projects live · 2 more in the 2026 roadmap
Each project includes
Working code on GitHub Architecture documentation Pluggable configuration
Project structure

Three artifacts per project, consistently.

Each project follows the same structure so you always know where to find what you need, whether you are evaluating the approach or running the code.

01
Overview
The business problem, how the system addresses it, what the output looks like, and what you need to run it.
02
Architecture
End-to-end pipeline diagram with every layer explained. Technical decisions documented with the reasoning behind each one.
03
GitHub
Modular Python code with a single configuration file. Clone the repository, set your API key, and run the notebook.
Impact

What these systems actually move.

Numbers pulled directly from each project's evaluation, not projections.

<10sec
Regulatory query time
Down from 20–40 minutes of manual search through FFIEC BSA/AML documentation. [GENAI-001]
1.19p.p.
Default risk reduction
Per one standard deviation increase in credit limit — the opposite of the common risk assumption. [CAUSAL-001]
4×lift
Retention targeting
Top-decile churn scores find up to 4 times more churners than random outreach. [RISK-001]

Project Index

3 projects · 2 in roadmap · Vol. 01 · 2026
ID Project Stack Status
[GENAI-001]
GenAI
BSA/AML Regulatory RAG
A retrieval-augmented generation system for querying BSA/AML regulatory documents. Answers compliance questions in plain English with citations down to section and page number. Uses hybrid semantic and keyword retrieval with temperature fixed at 0.0 for deterministic output.
Overview Architecture GitHub
LangChain
Gemini 2.0
ChromaDB · BM25
Live
[RISK-001]
Risk
Customer Churn with Explainability
A churn prediction pipeline for banking customers using gradient boosting with SHAP-based explainability. Each prediction comes with the features that drove it, making the output interpretable for risk committees and audit teams.
Overview Architecture GitHub
XGBoost
Optuna
SHAP
Live
[CAUSAL-001]
Causal Inference
Credit Intervention Analysis
A causal inference framework for estimating the effect of credit limit changes on default rates. Built with DoWhy and EconML to separate correlation from causation and answer the counterfactual questions risk teams actually need.
Overview Architecture GitHub
DoWhy
EconML
Causal Forest
Live
Roadmap 2026
[FCST-001] Transaction Volume Forecasting
[ENG-001] Data Engineering Foundations
About
Odemir Depieri Jr
Data and AI Specialist

14 years of experience in data and AI within banks and financial institutions, across the full lifecycle: architecture, engineering, and production deployment, focused on solving real business problems.

This hub is where I open source that work. Each project ships with the code, the architecture, and the reasoning behind it, no signup required.

GitHub LinkedIn
Areas covered in this hub
GenAI RAG systems, LLM pipelines, regulatory document querying
Risk Credit scoring, churn prediction, explainability for audit
Causal Inference Treatment effects, intervention analysis, counterfactual modeling
Platform MLflow, vector search, production deployment
Stack Python, LangChain, Gemini, XGBoost, DoWhy, EconML
Built with
Python LangChain Gemini XGBoost ChromaDB dowhy