Building systems for retrieval, security, market analysis, and scientific visualization.
I'm a software developer and founder of Directa. I work across systems programming, applied AI, and web development, connecting performance-sensitive components to APIs and interfaces people can use.
My projects explore how systems work beneath the surface: vector indexing, dependency reachability, market microstructure, multimodal signal processing, and geometric collision evaluation. I document architectural decisions, build reproducible demos, and make implementation limits explicit.
| Project | What it does | Stack |
|---|---|---|
| RAG Engine | HNSW vector index implemented from scratch, with Python bindings, a retrieval API, a source-citing chat interface, and reproducible benchmarks against brute-force search. | Rust PyO3 Python FastAPI Next.js |
| Vulnerability Triage Agent | Prioritizes dependency CVEs using structural reachability analysis, lockfile graphs, and OSV advisories. Includes offline triage, HTML reports, SARIF export, and GitHub issue drafting. | Rust Python TypeScript tree-sitter |
| Self-Healing Incident Agent | Incident-response demo connecting log and metric detection to diagnosis, proposed diffs, test validation, and a pull-request workflow with human review. Includes a reproducible offline mode. | Go Python SQLite Next.js |
| Trading Microstructure Engine | Processes tick data into order-flow imbalance, realized volatility, volume, and VWAP. Includes a trade journal, behavioral-analysis integration, dashboard, and one-command synthetic-data demo. | Rust Python FastAPI SQLite Next.js |
| Multimodal Biosignal Classifier | Educational activity-recognition system combining ECG, PPG, and accelerometer signals, with subject-separated evaluation, calibrated confidence, a prediction API, and signal visualization. | Python PyTorch FastAPI Next.js Docker |
| Occlusion Lab | Educational sandbox for synthetic dental geometry, contact visualization, and deterministic motion sweeps. In development: migrating toward a C++20 native collision and evaluation architecture. | C++20 CMake FCL TypeScript Three.js Rapier |
| Project | Focus |
|---|---|
| Polymarket PnL Forensic Auditor | Read-only wallet analysis: realized PnL, settlement accounting, on-chain maker/taker reconstruction, and selection-bias testing. Built with TypeScript, Node.js, and SQLite. |
| EDGE Predict Model | In development: sports decision-support foundations covering no-vig odds, expected value, fractional Kelly, and auditable picks. Quantitative core implemented; predictive modeling and backtesting remain pending. |
| Personal Portfolio | Bilingual portfolio with an editorial layout, localized content, and motion design. Built with Next.js, TypeScript, next-intl, and GSAP. |
- Build a working end-to-end slice before expanding scope.
- Keep numerical calculations and validation separate from model-generated explanations.
- Test system boundaries, failure paths, and data integrity.
- Document tradeoffs and the reasoning behind architectural decisions.
- Label synthetic data, offline modes, and unfinished features clearly.
Languages: Rust · Python · TypeScript · Go · C++
APIs & data: FastAPI · PostgreSQL · SQLite · PyO3
ML & analysis: PyTorch · HNSW · tree-sitter
Interfaces & delivery: Next.js · Three.js · Docker · CMake · GitHub Actions
Open to software engineering opportunities and collaborations in applied AI, security tooling, and data-intensive systems.