Just a Human In The Loop

• Building autonomous agents so they can take over the boring tasks (and maybe the world someday)
• Deploying production AI systems at Debales — real clients, real stakes, way less room for "oops, wrong API key"
• Building Warden — a pre-action authorization layer for AI agents, because right now the average agent fails ~57% of multi-step tasks and nobody's stopping it *before* it does something dumb with your production database
• Building real-time voice AI systems — (LiveKit, Deepgram, Cartesia) — teaching machines to pick up the phone so I don't have to
• Working on an autonomous stock trading environment — teaching AI to lose money faster than I can
• Crafting AI workflows that pretend to understand what I actually want
• Fine-tuning LLMs — because the base model clearly didn't get my sarcasm

• Generative AI Frameworks: LangChain, Llama-Index, HuggingFace — teaching machines to talk so they can ignore me better
• Agentic AI Frameworks: LangGraph, OpenAI Agents SDK, CrewAI, AutoGen, MCP — giving AI just enough freedom to cause chaos
• LLM Engineering: RAG, Fine-Tuning, Context Engineering — because apparently, telling AI what I want is rocket science
• Voice AI and Real-Time Systems — LiveKit, Deepgram, Cartesia — teaching machines to answer the phone so I don't have to
• AI Observability: LangSmith, LangFuse, OpenTelemetry — so when the agent inevitably breaks, I at least know why
• Cloud and Deployment — AWS (Bedrock, SageMaker), GCP, Azure, Vercel, Docker, Kubernetes — shipping this stuff somewhere that isn't my laptop
• Machine Learning & Deep Learning Algorithms — the wizardry behind all my AI "accidents"
• Natural Language Processing — making computers pretend they get human nonsense
• Data Structures & Algorithms in C++ — because sometimes you need to make things needlessly complicated
• Almost all core Engineering subjects — a lifetime subscription to "I survived engineering school" therapy

• Shipped a real-time voice AI system at Debales using LiveKit, Deepgram, and Cartesia — it answers calls, handles escalations, and transfers to humans without ever needing a coffee break.
• Built AI agents for logistics and e-commerce clients across multiple platforms and integrations.
• MAITRI AI – Intelligent Psychological Companion (Smart India Hackathon): Led a 5-member AI crew building a space-ready psychological companion that reads astronauts’ voices and faces in real-time, detects emotions with 92% accuracy, and responds naturally — basically an AI therapist keeping space crews sane while boosting mission efficiency by 35%.
• BioSage – AI Knowledge Engine (NASA Space Apps 2025): Built a multi-agent AI knowledge engine that ingested 600+ space-biology papers and all NASA official domains, delivered 98% citation accuracy, and let researchers pull insights 80% faster
• Built a RAG-based chatbot deployed on my portfolio site to answer recruiters' and stalkers' questions — and promptly snitch if they get weird
• Created a Multi-Agent Deep Research System capable of parallel information retrieval, reflection-based reasoning, and automated generation of structured research reports with improved coherence and depth. — because one AI just wasn't enough chaos
• Assembled a multi-agent AI engineering team with a lead, frontend, backend, and tester — all working tirelessly to do my bidding
• Fine-Tuned Llama-3.1-8B for predicting Product prices by their descriptions, with Mean Error of $47, outperforming traditional ML techniques for the same.
• Built a Self-Organizing AI Idea Refinement Ecosystem - an army of AI agents arguing with each other until they ‘accidentally’ invent something smarter than me. Zero human intervention, maximum existential dread.
• Developed an AI agent that spawns other AI agents who argue and collaborate to optimize ideas — basically my own little AI soap opera
• Built an autonomous SDR that writes and sends cold emails — so I can ghost people professionally
• Completed a hell lot of toy ML and deep learning projects — but don't let that fool you, I've got concepts down cold

• I want to do something for social good — because saving the world looks better on LinkedIn
• Keep shipping production AI at scale — turns out commercial AI pays better than my imaginary stock trading bot
• Land a role as an AI Engineer or Forward-Deployed AI Engineer— less "pretending to know what I'm doing," more actually doing it, on someone's production systems instead of just mine.