A showcase of my work in Software Development and Artificial Intelligence. Each project demonstrates practical applications of cutting-edge technology.
A free-data replication of arXiv:2607.05291 benchmarking Chronos-Bolt and IBM Granite TTM against HAR and GARCH across 30 assets and true 5-minute crypto RV. The answer is mostly no, and the write-up says so.
A Lambda-architecture platform over GitHub Archive that predicts which new repositories reach 1,000 stars within 90 days, with a batch Spark ML path and a live Kafka streaming path feeding one Grafana view.
A LangGraph trading agent on Base Sepolia whose limits are enforced by a smart contract rather than by its own prompt, with every decision's reasoning pinned to IPFS and readable from the dashboard.
Sales calls in, structured pipeline intelligence out — a Whisper to GPT-4o event chain extracting objections, buying signals and risks, behind strict per-organisation isolation.
Multimodal Gemini Live agent that sees NYC through your camera, narrates live, and grounds answers in real city data (311, restaurant inspections, crime stats) for that exact location.
Document-grounded Q&A for financial advisors — hybrid dense and lexical retrieval fused with RRF, a LangGraph agent that retries when its own answer scores poorly, and PII redaction before anything is indexed.
An anomaly detector taken through the whole lifecycle — ingest, train, track in MLflow, serve over FastAPI, monitor for drift with Evidently, and ship through CI to a container registry.
A real-time financial analytics platform integrating 7+ external APIs to deliver stock market data, economic indicators, cryptocurrency tracking, and AI-powered insights.
Runs adversarial prompt suites against a target model across five attack categories, scores each response, and streams progress live — with the scoring method named honestly rather than dressed up as an LLM judge.
A small Go service that concurrently health-checks HTTP endpoints, persists results to Postgres, and asks a locally hosted GGUF model to explain what the numbers mean, falling back to rule-based insights when it is unavailable.
Real-time multimodal gaming controller using MediaPipe for hand tracking, custom MLP classifier for gesture recognition, OpenAI Whisper and Vosk models for voice commands, and Groq API for ultra-low-latency LLM inference.