Open to ML, AI engineering, MLOps/DevOps, and cloud roles
ML · AI Engineering · MLOps/DevOps · Cloud
Builds ML and AI engineering systems for production. Personal portfolio of a machine learning engineer building ML, AI engineering, MLOps/DevOps, and cloud platforms that turn research-grade models into monitored, secure, deployable systems.
I work across the full stack of applied ML and AI engineering: data and streaming systems, model serving, evaluation harnesses, security gates, observability, and cloud deployment. My projects are built like production systems, not demos: containerized services, reproducible ML workflows, and CI pipelines that test both behavior and safety.
Experience
Reduced routing errors by 50% with a FastAPI smart router using LLM intent classification and state-preserved task frames; improved system resilience by 60% with reverse proxy security, rate limiting, circuit breakers, queues, and fail-closed Stripe entitlement checks.
Reduced manual compliance lookups by 80% and automated 70% of manual verification steps by building a production platform for a Government IPR initiative with natural-language search and containerized CI/CD.
Saved $55K+ annually with a 96% infrastructure cost reduction by consolidating NAT Gateways via Terraform; cut release cycles by 40% with Jenkins to ArgoCD GitOps and Trivy/SonarQube quality gates.
Selected Projects
Production-grade autonomous drone fleet AI platform combining real-time telemetry, GPU vision inference, stream analytics, LangGraph mission coordination, MCP policy gates, and DevSecOps automation.
Cut manual agent evaluation overhead by 4x and increased throughput by 3x with a FastAPI + WebSocket orchestration server that models Terminal-Bench tasks as an OpenEnv-compliant RL environment.
Secured 93%+ processing accuracy at a sub-$0.01/document cost target by building a containerized OCR ensemble with selective SLM/VLM adjudication and isolated Docker microservices.
Reduced unnecessary LLM usage with a 3-tier confidence cascade and improved response trust using PII scrubbing, HyDE, BM25 + FAISS, cross-encoder reranking, abstention gates, and citation enforcement.
Publication
Zenodo Record · 2026
This paper proposes an RL exploration method that adapts intrinsic curiosity rewards over time and uses episodic memory to reduce repeated exploration of already-visited states. The approach targets sparse-reward environments where agents need a stronger signal for novelty, memory, and long-horizon discovery.
Toolbox
AWS, Kubernetes, Docker, Terraform, ArgoCD, Jenkins, GitHub Actions
MLflow, DVC, Ray Serve, Redpanda, Apache Flink, LangGraph, LangSmith, MCP
Trivy, OWASP ZAP, SonarQube, Falco, Python, Bash, YAML, PyTorch
Recognition
Semi-Finalist · Pan-IIT AI/ML Hackathon powered by IDFC First Bank, 2026
National Rank 4 · Launch Pad 2025, Pan-India Case Competition at SRCC
Contact