AI Engineer focused on taking systems from prototype to production. I architect multi-agent orchestration pipelines, low-latency FastAPI services, and vector-search workflows that ship to real users. Currently building the foundational AI stack at Medikabazaar.

• Agentic AI: robust memory layers, multi-agent orchestration with LangGraph, tool-driven autonomy.
• Retrieval at Scale: Qdrant / Pinecone / FAISS / ChromaDB pipelines tuned for latency and cost.
• Production Engineering: FastAPI + async + queues (TaskIQ / Celery / RabbitMQ) on AWS & Vertex AI.
• LLM Infrastructure: AI Command Center - a zero-dependency LLM gateway + self-hosted cost/usage dashboard, live and open-source.
• Published Packages: ai-command-center & @ai-command-center/sdk on npm; asterix-agent, oscar-agent, qmem & aicc-sdk on PyPI.

(February 2026 – Present)  ·  Founding member of the AI initiative

• Shipped the Image Enhancement Platform (flagship): FastAPI service deployed on AWS (Vertex AI, S3, Kafka, EC2, Jenkins CI) - ~14,000 product images live on the marketplace, all converted to WebP for measurable SEO + page-load wins
• Built the SEO Blog Agent end-to-end: Python pipeline ingesting Ahrefs keywords against live SKUs, clustering them into 624 product groups, plus a React/TypeScript dashboard, automated three-size banner generator, and CMS publishing - blogs already live with zero engineering involvement per release
• Architected the AI Command Center (internal "AI Box" initiative): a 6-layer hybrid multi-agent platform (orchestrator + 5 shared functional agents + per-product liaisons) for company-wide observability, evaluation, cost tracking, and guardrails - projected 40–60% reduction in org-wide LLM spend; independently built the core end-to-end and open-sourced it as 3 packages across npm + PyPI (repo · live site)
• Delivered audits across 27 codebases + JIRA (now a recurring monthly responsibility) and ~₹3L+ in direct cost savings via Gemini Batch API, async pipelines, cache-hit optimisation, and a JSpreadsheet → community-edition migration
• Drove AI enablement across the org: onboarded engineers and non-technical teams on Claude / Claude Code / Antigravity / Amazon Q; integrated Google Ads, Analytics, and Search Console with Claude Code via MCP

Stack: Python FastAPI AWS Vertex AI Kafka S3 EC2 Jenkins Gemini API React TypeScript Playwright Firecrawl

(June 2025 – October 2025)

• Built FastAPI backend integrating Google Gemini API and OpenAI embeddings for legal document analysis, semantic search, and text-to-SQL generation on DuckDB
• Designed vector embedding workflows with Qdrant and Redis caching, reducing query latency by ~90%
• Developed multi-agent HR automation system using LangGraph and MongoDB for candidate evaluation and behavioral analysis
• Implemented async data pipelines with TaskIQ/Celery and RabbitMQ for background assessment generation
• Created QMem, a Python CLI library for vector search automation (published on PyPI)

Stack: Python FastAPI LangGraph Qdrant MongoDB Redis RabbitMQ Gemini API DuckDB

(February 2025 – April 2025)

• Developed and fine-tuned 15+ ML/DL models (XGBoost, CNNs, Neural Networks) for classification and regression tasks on 100K+ row datasets
• Created reusable preprocessing pipelines from scratch across tabular, time-series, and image/text data
• Deployed 10+ trained models into production using Streamlit, Gradio, FastAPI, and Django

Stack: Python TensorFlow XGBoost Streamlit FastAPI Django Pandas NumPy

Point any project - in any language - at one base URL, and every token, rupee, and millisecond lands in one self-hosted dashboard. Designed as the internal "AI Box" platform, then built end-to-end and open-sourced as 3 packages across npm + PyPI.

• Core Tech: Node.js (ESM, zero runtime deps), HTTP reverse proxy + SSE streaming, Vanilla JS dashboard, Next.js docs site on Vercel, Python + JavaScript SDKs
• What's inside: 10+ LLM providers through one gateway (OpenAI, Anthropic, Gemini, Mistral, Groq, Ollama + any OpenAI-compatible), exact per-request cost accounting in multi-currency (INR/USD/EUR), provider routing/failover, RBAC (3 roles + per-project keys), budgets + anomaly alerts, session traces, PII-safe (no message bodies stored)
• Engineering: <0.25 ms p50 proxy overhead · 0 runtime dependencies in the core · 70+ tests with GitHub Actions CI across Node 18/20/22 · 3 published packages (ai-command-center · @ai-command-center/sdk · aicc-sdk)
• Why it matters: One language-agnostic control plane for cost, guardrails, and observability across an org's entire AI portfolio - sole designer & developer, built during my internship at a healthcare B2B

Python library for stateful AI agents with editable memory blocks and semantic retrieval. Published on PyPI.

• Core Tech: Python, Qdrant Cloud, SQLite, decorator-driven tool registration
• Why it matters: Lets agents persist context across sessions and retrieve memory semantically - the missing piece for production-grade autonomous systems

VS Code extension + CLI powered by my own Asterix framework and Gemini 2.5 Flash via Vertex AI, specialized for git workflows - branch comparison, PR review, diff analysis, and safe automation.

• Core Tech: Python, FastAPI (SSE streaming), TypeScript VS Code extension, Asterix (ReAct loop + memory), Vertex AI, Playwright, Tavily
• What's inside: 15 registered tools (9 git · shell · web search · browser), 12 HTTP endpoints, 4-tier risk model with human-in-the-loop confirmations, typed CONFIRM for dangerous ops, rotating JSONL audit log
• Why it matters: Real demonstration of agentic orchestration with production safety: ~2.3K backend LOC, 15 tests, CI across Python 3.10/3.11/3.12, ~9 months of iteration

Real-time SQL autocompletion inside MySQL Workbench & PostgreSQL with 95%+ acceptance rate across 20+ sessions.

• Core Tech: LLaMA 3 via Groq, Pinecone, MLflow, Docker, YAML-driven configs
• Impact: Latency cut by 50%+ via Dockerized deployment with preloaded models and clipboard hooks

Multi-PDF semantic search with OCR and cross-document theme synthesis.

• Core Tech: FastAPI, FAISS, Tesseract OCR, Sentence-Transformers, Groq LLM
• Deployed: Hugging Face Spaces (backend) + Vercel (frontend)

Intelligent tutoring system delivering personalized learning paths through multi-agent orchestration.

• Core Tech: Python, LangGraph, FastAPI, ChromaDB, OpenAI, Gemini
• Why it matters: Democratizes access to high-quality, personalized education by adapting to each learner's pace and style