Leonardo Barretti
AI / LLM Engineer — Production LLM Systems
I design and build production-grade AI systems — multi-agent pipelines, RAG, and self-correcting workflows with real observability and cost control. 15+ years as a Business Systems Analyst before transitioning into applied AI.
Featured Projects
Cronograph
Self-Hosted LiveProduction platform that extracts high-resolution market data from Binance and runs statistical window analysis to support weekly Bitcoin options strike selection. Sub-50ms aggregation over hundreds of thousands of OHLCV candles.
AuditChain
Self-Hosted Live100% recall and 50% precision on a curated 7-case eval set (Bausch Health/Valeant, WorldCom, Luckin Coffee + 4 clean controls). Calibrated to prioritize fraud detection over false-positive minimization — the right tradeoff for forensic auditing.
bitPredict
Self-Hosted LiveMulti-timeframe Bitcoin forecasting powered by Kronos, a 102M-parameter foundation model (HuggingFace). 30 stochastic simulations per candle with calibrated uncertainty bands and a portfolio backtest engine reporting Sharpe, drawdown, and win rate.
1.About Me
I'm an AI Engineer with an unusual background: 15+ years as a Business Systems Analyst in Brazilian enterprises — XP Investimentos, Petrobras, Accenture — followed by a deliberate transition into applied AI.
That long detour through requirements gathering, stakeholder communication and project delivery shapes how I build today: I'm wary of LLM hallucinations in numeric outputs (so I designed deterministic risk scoring in AuditChain), I instrument systems to know exactly what each agent costs and how long it takes (custom observability across all 4 portfolio projects), and I test changes against curated eval sets instead of vibes.
I care about reliability, cost per inference, and being honest about what's measured versus what's assumed.
Open to opportunities in AI Engineering.
2.Tech Stack
Other Projects
RAG Systems
Retrieval-augmented generation pipelines with pgvector, citations, and streaming responses.
Doc Pipelines
Document-processing pipelines with LLM extraction, classification, and structured output.
Workflow Automation
AI-driven workflow automation for international clients — requirements to deployed products.