Building Autonomous Agentic Workflows, Production GenAI Backends (FastAPI & pgvector), and Edge Machine Learning Systems.
I build production-grade AI systems end-to-end — from asynchronous backends and vector search engines to autonomous multi-agent control loops and edge ML deployments. My technical core centers on Python, FastAPI, pgvector, LangChain/ReAct architectures, PyTorch, and TensorFlow Lite. I focus on measurable engineering impact: reducing API latency, eliminating hallucination loops, optimizing token budgets, and shipping clean, maintainable systems.
DEGREE
B.E. in AI & ML (CGPA: 7.77)
EXPERIENCE
2 Machine Learning & Android Internships
SHIPPED
8 End-to-End AI & Full-Stack Systems
Kubernetes Multi-Agent AI Execution Platform engineered for asynchronous job queuing, Terraform IaC, GitOps delivery, and full-stack telemetry.
Running multi-agent AI tasks synchronously causes gateway timeouts, lacks workload isolation, and lacks cluster execution metrics.
Engineered an async task engine on AWS EKS using FastAPI, Celery, and Redis, automated GitOps delivery via Argo CD, and integrated Prometheus/Grafana telemetry.
Full-Stack AI Career Intelligence Platform featuring multi-provider LLM orchestration, async API throughput, and pgvector semantic resume search.
ATS keyword matching misses contextual candidate experience and codebase quality, delaying talent discovery.
Built an async FastAPI backend ingesting Resume PDF, GitHub repository histories, and LinkedIn profiles into a unified candidate representation with hybrid pgvector search (<200ms API response).
Autonomous ReAct agent control loop built from scratch featuring deterministic evaluation, budget-capped recovery engines, and prompt reformulation.
Standard LLM tool-calling fails silently or enters infinite loops upon malformed tool responses or missing API parameters.
Engineered a multi-step ReAct framework from scratch featuring modular components: Planner, Orchestrator, Tool Router, Working Memory, and Evaluator with budget-capped state backtracking.
Resilient multi-source ETL pipeline and tiered entity resolution engine for customer record reconciliation.
Inconsistent customer records spread across legacy databases, rate-limited APIs, and varied CSV encodings generate duplicate data.
Engineered an idempotent ETL pipeline with tiered entity resolution combining exact key matching with C++ accelerated RapidFuzz fuzzy similarity scoring to generate a golden record.
Generative AI community mentorship mobile application with on-device voice assistance and Google Gemini 2.0 Flash mentor matching.
Multilingual offline-first cattle disease diagnostic mobile app powered by INT8 quantized TensorFlow Lite edge inference.
Asynchronous multi-agent job discovery engine running background scraping, ChromaDB match scoring, and automated outreach generation.
Interactive web dashboard for visualizing CNN feature maps, filter weights, and activation layers for model interpretability.
Civic engagement app connecting citizens with ward representatives using Firebase Auth and Supabase Row Level Security.
Srinivas Institute of Technology, Mangalore