Below is a selection of projects developed under agile methodologies, focused on solving business problems through Generative AI and data analytics.
Generative AI & Agents
Intelligent Help Desk (CENACE)
Technical Support Optimization with Private AI
System designed for the National Energy Control Center. It solves the need to automate incident classification and suggest technical solutions without exposing sensitive data to the public cloud.
Key Points:
- Implementation of Open-Source models (Gemma/Ollama) for total data privacy.
- RAG architecture for querying internal technical manuals.
- Persistence of history and sessions in MongoDB.
Product Recommender (CT)
Real-Time Sales and Recommendation System
Advanced chatbot for hardware suggestions. Overcomes the challenge of connecting a massive MySQL catalog with a semantic search engine, managing dynamic prices and stock.
Key Points:
- Complex ETL pipeline to synchronize MySQL with Vector Stores (FAISS).
- “Mixed Content” architecture solution via PHP Proxy.
- User moderation system and custom JS widget.
Financial Chatbot (Salsas Castillo)
Sales and Financial Reports Assistant
Conversational solution deployed on Telegram for executives and the sales force. Allows natural language queries over transactional databases and internal regulations.
Key Points:
- Orchestration of SQL Tools for real-time sales queries.
- Automated generation of PDF Reports with AI analysis.
- Native integration with Telegram API and audio transcription (Whisper).
Data Engineering
ETL & Data Engineering (Frutal)
Data Architecture Modernization
Design of a robust pipeline to migrate a legacy ERP system (HFSQL) to a modern Data Lake, resolving critical driver incompatibilities.
Key Points:
- Solution to the “Stack Smashing” error by isolating processes in Python.
- Containerization with Podman to unify development (Windows) and production (Linux) environments.
- Storage and read optimization by converting to Parquet format.
Data Science & Analytics
Customer Analytics (CT)
From Analysis to Production
Customer Analytics platform built for CT Internacional. It covers everything from RFM segmentation and identifying association rules (Market Basket), to the deployment of a daily ETL pipeline and an interactive dashboard.
Key Points:
- K-Means (Segmentation) and FP-Growth (Market Basket) models.
- Modular ETL pipeline persisted in Parquet.
- Operational dashboard built with FastAPI, DuckDB, and Plotly.