Headline

Full Stack & AI/ML Engineer | National Finalist- KAVACH Cyber Security Hackathon 2023| GenAI & Agentic AI | LLMs, RAG & LangGraph | MCP | Python | FastAPI | Multi-Agent | Node.js | Prompt Engineering | Docker & AWS

About

𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 𝐢𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞, 𝐞𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐜𝐲, 𝐚𝐧𝐝 𝐞𝐧𝐭𝐞𝐫𝐩𝐫𝐢𝐬𝐞 𝐢𝐦𝐩𝐚𝐜𝐭 - 𝐭𝐡𝐚𝐭’𝐬 𝐰𝐡𝐚𝐭 𝐝𝐫𝐢𝐯𝐞𝐬 𝐦𝐞 With 2+ years of experience across AI/ML and full-stack engineering, I build intelligent systems that simplify complexity, automate manual work, and speed up decisions. My work spans agentic AI, retrieval-augmented generation, and production backends deployed on AWS. At Car Wizard (VahanHelp), I owned the AI layer of a live internal CRM used by operations and support teams to process 10,000+ vehicle-transfer cases for 8,000+ customers and dealers. I built semantic case search on pgvector that cut lookups from 5–10 minutes to under a second, shipped a RAG policy assistant served through ReAct agents, LangGraph state machines and MCP tools, and lifted end-to-end RAGAS scores from 0.60 to 0.93 with better chunking and cross-encoder reranking. Earlier, at DigiLocker, I built a facial anti-spoofing system that reached 96.4% classification accuracy and an EER under 2.5%. I pair that AI work with full-stack ownership: async FastAPI services, React/TypeScript frontends, and a live React Native app. Everything ships behind eval harnesses and shadow-mode gates, so the systems are measurable, auditable, and safe to trust. 𝐌𝐲 𝐬𝐭𝐫𝐞𝐧𝐠𝐭𝐡𝐬: - Agentic AI: LangGraph, LangChain, ReAct loops, MCP, tool calling, structured outputs - RAG & Retrieval: hybrid search (dense + BM25), chunking, cross-encoder reranking, pgvector - LLM Integration: Claude, OpenAI, prompt engineering, guardrails, human-in-the-loop - Evals & Observability: RAGAS, DeepEval, LangSmith, gold datasets - ML & Computer Vision: PyTorch, Hugging Face, fine-tuning - Backend & Full Stack: Python, FastAPI, Node.js, TypeScript, React, React Native, PostgreSQL, MongoDB - Cloud & Deployment: AWS (EC2, S3, Lambda, RDS, SageMaker), GCP, Docker, Kubernetes 𝐇𝐞𝐫𝐞’𝐬 𝐰𝐡𝐚𝐭 𝐈 𝐛𝐫𝐢𝐧𝐠 𝐭𝐨 𝐭𝐡𝐞 𝐭𝐚𝐛𝐥𝐞: - System Thinking: Architectures that connect data, retrieval, agents, and business workflows - End-to-End Ownership: From embeddings and evals to APIs, apps, and deployment - Execution Rigor: Prototype to production, measured rather than guessed Side projects: an email and calendar agent with prompt-injection guardrails and approval gates, and QueryPilot, a text-to-SQL copilot that narrows a 500-table schema with RAG. National Finalist, KAVACH Cybersecurity Hackathon 2023. CodeChef 1900+. Let’s connect if you’re building with LLMs, RAG, or AI agents. I’m open to AI/ML Engineer, LLM/Agentic AI, and AI Backend roles.

Location

Noida

AI stack

antigravity

claude code

claude

codex

cursor

geminicli

Work experience

Software Engineer at VAHANHELP, May 2024 - Present · 2 yrs 5 mos, Noida, - Built the full admin/agent dashboard for a multi-department CRM using Angular 17+ (standalone components) on the frontend and Node.js/Express, MongoDB on the backend, with RBAC and aggregation-pipeline reporting. Implemented real-time features via WebSockets — live case/lead status updates and agent-to-agent chat — so teams could coordinate without manual refreshes or status calls. Owned end-to-end architecture and AWS deployment, including a companion React Native app (Play Store) with OTP auth and Razorpay integration. - Built a semantic case-similarity search feature for a production RC-transfer case database: embedded case notes with OpenAI's text-embedding-3-small into PostgreSQL/pgvector, exposed via FastAPI so agents retrieve similar past cases instantly instead of manual search - Designed a RAG-based policy assistant giving support staff conversational access to RTO transfer rules and SOPs, served through an MCP Python server exposing case-search and policy-lookup as callable tools — deployed on AWS EC2/Docker - Re-architected a bulk vehicle-lookup workflow from 200 sequential API calls into 20 concurrency-controlled batch calls with partial-failure handling and exponential backoff -cut processing time from 4 minutes to under 25 seconds - Designed backend services and schema for a multi-department CRM (Node.js/Express, MongoDB) with RBAC and aggregation-pipeline reporting; owned end-to-end architecture and AWS deployment of a live Play Store app (React Native) with OTP auth and Razorpay integration

AI/ML Intern at DigiLocker, Mar 2024 - May 2024 · 3 mos, Remote, During my internship at DigiLocker, I specialized in AI and ML, focusing on enhancing digital document security. I developed a classifier using transfer learning to distinguish real from spoofed facial images with high accuracy. Additionally, I engineered a facial antispoofing system from scratch, significantly improving detection accuracy. Implementing dual-technique approaches based on research papers, I contributed to creating a robust facial antispoofing mechanism, refining my skills in AI, ML, and cybersecurity.

Education

Bachelor of Technology - BTech, Dr. A.P.J. Abdul Kalam Technical University (AKTU), Lucknow