Built a rolling-baseline anomaly detector and cross-slice root-cause analyzer that isolates payment-gateway degradation (e.g. UPI/PayFast outages) from healthy traffic, then routes affected transactions to an LLM-driven recovery agent (OpenAI gpt-40-mini) that reasons over context and chooses a bounded action (retry, delay, discount, reroute, escalate). Enforced deterministic guardrails (retry caps, cooldowns, discount limits, fraud-pattern escalation, idempotent execution) around the LLM's output; benchmarked the agent against a naive baseline across five independently detected degradation events, recovering 82.29% of degradation-linked revenue vs. 30.73% for the baseline, backed by a pytest/GitHub Actions CI suite.

Designed a serverless microservices workflow using FastAPI endpoints on AWS Lambda to ingest health metrics and analyze failure propagation across dependent services; orchestrated metric ingestion and analysis triggers using Amazon EventBridge. Implemented directed graph traversal for cascade-path detection and modeled DynamoDB time-series storage with CloudWatch-based latency and execution monitoring; provisioned infrastructure using AWS CDK.

Built an Express.js API with JWT-based authentication (Bearer tokens) and basic request validation; integrated MongoDB (Mongoose) for query/feedback persistence and Redis connectivity with a health check endpoint. Built a React + TypeScript UI for citation-based answers with an evidence viewer, plus query history and feedback; containerized the app using Docker for reproducible local development.

CGPA: 8.61/10

Class XII: 87.5% — Class X: Pass (State board)