Node.js Express.js MongoDB Redis React Socket.IO Jest

• Diagnosed a ~2.5s median message latency under cross-region load (Render Oregon ↔ MongoDB Atlas Mumbai / Upstash Redis).
• Fixed it with Redis cache-aside validation and fire-and-forget cache invalidation — 81% latency cut, down to ~468ms.
• Solved message-ordering race conditions with server-side timestamps; built accurate seen-status tracking via Socket.IO acknowledgements.
• Isolated Jest integration suite (mongodb-memory-server) covering auth and messaging routes, including error paths.

Repo · Live Demo

MERN Gemini OpenRouter Groq Hugging Face

• Built a multi-provider LLM routing system with circuit-breaker fallback across four providers.
• Load-tested against real provider failures (rate limits, timeouts) — 100% request success via automatic failover.
• Groq as primary provider, ~780ms median latency, roughly 3x faster than the Hugging Face fallback.
• Adaptive interview difficulty based on live user performance; resume parsing (PDF/Text) for personalized scenarios; automated PDF feedback reports.

Repo · Live Demo

⚠️ Repo name pending confirmation — verify this matches your actual GitHub repo before publishing.

Python SQL Power BI Random Forest

• Random Forest classifier with class-weight balancing for imbalanced delay labels — 87% accuracy, 18 points above the logistic regression baseline (69%).
• 93% precision on on-time order predictions.
• Full ETL pipeline: SQL ingestion → cleaning → feature engineering + one-hot encoding → model training → Power BI dashboard.
• SQL feature engineering using multi-table joins, CASE WHEN, GROUP BY/HAVING.

Repo

No live dashboard link exists in your source resumes for this project — add one if you have it.

Python Streamlit SQL DuckDB Plotly

• Warehoused 1,200+ matches and 294K+ deliveries into DuckDB via modular Python scripts.
• Season trends, venue par scores, toss impact, phase-wise team/player performance via interactive Plotly charts.
• SQL with CTEs and window functions (QUALIFY ROW_NUMBER() OVER (PARTITION BY...)) for batter rankings.
• Supporting ML models for win probability and score projection, deployed to a production Streamlit dashboard.

Repo · Live App

Python Pandas NumPy Streamlit

• Cleaned and analyzed 10,000+ NCRB crime records; year-over-year trend analysis and state-wise comparison.
• Production Streamlit dashboard with choropleth maps, crime distribution charts, state-wise filters.
• Growth-rate forecasting projecting 2021–2023 trends per state; per-capita normalization for fair cross-state comparison.

Repo · Live App