Shriya Bansal

Automated pipelines ingesting 20 years of S&P 500 data (5,030 trading days) into 1,000+ weekly observations. Calculated an annualized 0.58 Sharpe ratio, rolling 52-week volatility, and return distributions from −18.2% to +12.1%. Includes a backtesting module and an analytical Matplotlib dashboard with Pearson correlation for asset hedging classification.

iOS & Android app designed for UTD students. Features a time-series risk profiler (conservative / moderate / volatile), dynamic leaderboard, and group financial pact creation with spending limits, deadlines, and member invites.

Local-first RAG system enabling natural language queries over user-uploaded PDFs — no third-party cloud exposure for sensitive finance and healthcare data. Tuned top-k similarity search (k=5) and 1,000-token chunk size. Event-driven pipeline with 3,072-dimensional vector embedding, semantic retrieval, and LLM response generation.

Surveyed and compared three LLM-based GPU kernel generation frameworks (ReGraphT, CUDAForge, TritonGym), analyzing their approaches to performance feedback, correctness verification, and iterative refinement. Proposed a closed-loop architecture combining MCGS-guided optimization with hardware profiling as a reward signal to address limitations in existing workflows.