• Architected a multi-agent GenAI platform (LangGraph, FastAPI) orchestrating 4 specialist agents with hybrid retrieval via ChromaDB vector search and Neo4j knowledge-graph traversal. • Engineered multi-modal ingestion (PDF, DOCX, XLSX, diagrams) using GPT-4 Vision and Tesseract OCR with SHA-256 deduplication. • Devised a 4-layer LLM safety system — groundedness verification, prompt-injection detection, high-stakes escalation, and PII redaction — with full audit logging.

• Engineered a microservices trading platform (Java 21, Spring Boot, React) with an order-matching engine using O(log n) priority queue DSA and TDD. • Applied enterprise design patterns (Strategy, Factory, Observer) to decouple limit/market order processing and real-time price updates. • Structured a PostgreSQL schema with PL/pgSQL triggers for transactional integrity; integrated a GenAI-powered LLM endpoint for portfolio insights.

• Co-authored a research paper extending HAR-RV models with FinBERT sentiment and a hybrid Neural-HAR architecture (FT-Transformer + GRN) across 3 asset classes (~8,300 observations). • Achieved statistically significant forecast improvements for all 3 assets (Diebold-Mariano, p < 0.05); validated via volatility-targeting backtest showing up to 7pp drawdown reduction. • Identified and fixed a data-alignment bug in the evaluation pipeline that had inverted a significance result, ensuring full reproducibility.