Final-year Software Engineering student — B.Tech Mining Engineering at NITK Surathkal + BS Data Science at IIT Madras (2027). Specializes in backend and distributed systems in Go and Python — job queues, semantic search engines, ML monitoring — with hands-on concurrency, REST/gRPC, and observability experience. Seeking full-time Software Engineer / Backend / Platform roles, remote-first preferred.
Built a VS Code chatbot powered by a local LLM (Ollama/LLaMA3) that classified security queries and routed them to SonarQube, Snyk, or Burp Suite for pre-push vulnerability detection with zero data egress. Architected a dual-mode scan engine on Node.js/Express with 18 REST endpoints and real-time streaming, surfacing 24 real vulnerabilities across 18 files on a live enterprise instance.
Built a real-time anomaly detection pipeline for streaming API and IoT sensor data, cutting false-positive alerts by 35%. Deployed ML inference on AWS EC2 with Docker, achieving sub-100ms latency with S3-backed model versioning.
Modeled directional anisotropy in blast-induced ground vibration using a stacking ensemble (XGBoost, SVR, Ridge), achieving R²=0.819 and 16-20% RMSE reduction over the Holmberg-Persson baseline. Findings in preparation for submission to Rock Mechanics and Rock Engineering (Springer, IF 6.2).
Redis-backed reliable job queue with goroutine worker pools, exponential-backoff retries, dead-letter queue, and idempotent enqueue. Crash-recovery via heartbeat leasing and reaper goroutine — zero job loss across 50 forced crash iterations and 1,000 concurrent jobs. REST API with Postgres audit trail and Prometheus metrics.
Natural-language code search using tree-sitter AST chunking at function/class boundaries. Two-stage retrieval — bi-encoder candidate generation over FAISS, cross-encoder re-ranking. FastAPI REST service + Typer CLI, metadata persisted via SQLAlchemy.
Replaced the traditional Holmberg-Persson formula with a per-component stacking ensemble, achieving R²=0.819 and a 4.14% RMSE improvement. SHAP-based directional feature importance analysis. Deployed as an interactive Streamlit predictor.
End-to-end causal pipeline recovering a true treatment effect (HR=0.75) from data exhibiting Simpson's Paradox (naive HR=1.35). Interactive dashboard with Kaplan-Meier curves, forest plots, and propensity diagnostics.
Go, Python, SQL, JavaScript, TypeScript
REST/gRPC APIs, goroutines & channels, concurrency patterns, Kafka, Redis, FastAPI, Node.js/Express
Docker, Kubernetes, AWS (EC2, S3), GitHub Actions, Prometheus, PostgreSQL, CI/CD
PyTorch, scikit-learn, XGBoost, HuggingFace, sentence-transformers, FAISS, LangChain, RAG, LoRA/QLoRA