Projects

Correlates logs, metrics, traces, and deployment events to rank likely incident causes with an evidence trail.

Architected an automated RCA system for distributed microservices by building an ingest + normalization pipeline, modeling service interactions as a temporal dependency graph, and applying causal scoring over time-ordered events. Used an LLM only for human-readable explanations while keeping inference deterministic and debuggable.

FastAPI service that turns job descriptions into structured, recruiter-ready application content.

Built a FastAPI backend to analyze job descriptions and return structured application outputs through clean REST endpoints. Optimized the workflow for fast iteration during active job searches and designed the architecture to be extensible for resume parsing, ATS optimization, and rule-based validation.

RAG system + evaluation framework for domain-adaptive telecom QA.

Built a telecom-focused RAG pipeline around LLaMA and retrieval tuning, then validated improvements using a structured evaluation framework and dashboards for quality monitoring.

Deployed LLaMA inference as scalable microservices with CI/CD and observability.

Containerized AI inference services and deployed them with autoscaling, blue/green releases, and production-grade monitoring and security controls.

Real-time gesture classification using live video landmarks.

Built a real-time computer vision pipeline that extracts hand landmarks from webcam streams and classifies gestures for human-computer interaction use cases.

Compared classical ML and sequence models for time-series prediction.

Developed a stock price prediction workflow using historical market data, feature engineering, and multiple model baselines to compare trend prediction performance.