I build production systems and measure them: an 81% latency cut on a real-time chat app, 100% uptime through a multi-provider LLM failover system, and full-stack apps shipped end to end on the MERN stack.
B.Tech Computer Science (IoT) graduate (2026) from Manipal University Jaipur (2022–2026, CGPA 8.41). I split my work between building full-stack products and analyzing data — and I care about the same thing in both: does it hold up under real load, and does the number actually move.
On the development side, I've diagnosed and fixed a cross-region latency bottleneck (81% reduction) in a production chat app, and built a fault-tolerant multi-provider LLM system that held 100% uptime through real API failures — rate limits, timeouts, all of it.
On the analysis side, I've built end-to-end ETL pipelines and dashboards across logistics, sports and public-safety data — including 294K+ ball-by-ball IPL records and 10,000+ NCRB crime records — and used supporting ML models (Random Forest, classification, regression) where prediction actually adds value, not as decoration.
Multi-provider LLM routing system (Gemini, OpenRouter, Groq, Hugging Face) with circuit-breaker fallback. Load-tested against real provider failures — 100% request success via automatic failover, Groq primary at ~780ms median latency. Adaptive question difficulty, resume parsing for personalized scenarios, voice-enabled practice, automated PDF reports.
Real-time chat app on Socket.IO with cursor-based pagination. Solved message-ordering race conditions via server-side timestamps and built accurate seen-status tracking. Diagnosed a ~2.5s median cross-region latency issue and fixed it with Redis cache-aside validation — 81% latency cut to ~468ms. Isolated Jest integration suite for auth and messaging routes.
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 orders. Full ETL: SQL ingestion → cleaning → feature engineering → training → Power BI dashboard.
End-to-end pipeline warehousing 1,200+ matches and 294K+ deliveries into DuckDB. Season trends, venue par scores, toss impact, phase-wise performance via Plotly. Window-function SQL (CTEs, QUALIFY ROW_NUMBER) for batter rankings; supporting ML for win probability and score projection, deployed to Streamlit.
Cleaned and analyzed 10,000+ NCRB crime records with year-over-year trend and state-wise comparison. Production Streamlit dashboard with choropleth maps and state filters. Growth-rate forecasting for 2021–2023 projections, per-capita normalization for fair cross-state comparison.