• AI Agent Engineer → Multi-agent orchestration, tool-calling, and agent safety rails.
• Full-Stack Developer → Next.js / React / TypeScript on the front, FastAPI & Node on the back.
• Applied ML → Data-centric model building, evaluation, and reproducible datasets.
• Infra & DevOps → CI/CD, observability, containerised deploys, incident automation.
• Protocols & Tooling → Model Context Protocol (MCP) servers, LLM gateways, agent runtimes.
• Open Source → Publishing datasets, demos, and agent tooling in the open.
• Currently → Computer Science undergrad, BITS Pilani.
I like building systems that actually run — agents that decompose real goals, call real tools, and ship verifiable results. Less demo, more production.
Every commit I've ever pushed, rendered as a skyline — rebuilt daily by github-profile-3d-contrib.
Night Rainbow |
Night Green |
Git Block |
Green |
Green — animated |
Seasons |
Seasons — animated |
Southern Seasons |
Southern Seasons — animated |
|
→ Adaption AutoScientist Challenge — data-centric AI submission:
- Built two original, self-verifying datasets and models: a function-calling model trained to know when not to call a tool, and a bilingual (English/Hindi) multimodal chart-reader.
- Published the datasets on Hugging Face and Kaggle, with a live demo on Hugging Face Spaces and Vercel.
→ CrewAI Multi-Agent Research System — production-grade agent pipeline:
- Four cooperating agents (Researcher → Analyst → Writer → Verifier) that turn a plain-language query into a cited, downloadable report, with a real-time dashboard tracking every stage.
- Hardened past MVP scope: job cancellation/retry, JWT revocation, per-report cost tracking, request tracing, Alembic migrations, and CI on every push.
→ OmniPotent — autonomous infrastructure intelligence platform:
- Multi-agent incident resolution with deterministic safety controls and deep observability integration, backed by CI and coverage reporting.
→ Open-source AI tooling:
- Shipped a remote Model Context Protocol (MCP) server, an autonomous triage agent, and experiments with LLM gateway routing across 1,600+ models.
| Project | Description | Repo & Demo Link |
|---|---|---|
| Adaption AI Lab → Data-Centric AutoScientist | Two models built around original, self-verifying datasets: a function-calling model that knows when not to call a tool, and an English/Hindi multimodal chart-reader. Datasets published to Hugging Face and Kaggle with a live inference demo. | 🔗 Repo · 🔗 Demo |
| CrewAI → Multi-Agent Research System | Four cooperating CrewAI agents (Researcher → Analyst → Writer → Verifier) turn a plain-language query into a cited, downloadable report, with a real-time dashboard to track every stage as it happens. Includes job cancellation/retry, JWT revocation, cost tracking and request tracing. | 🔗 Repo |
| Proxim → AI Conversation Training Lab | Browser-based lab where pharma sales reps rehearse live conversations with AI-simulated healthcare professionals. Streamed dialogue and ElevenLabs speech drive viseme keyframes on a Three.js 3D avatar, keeping voice and lip motion tightly coupled — no native binaries, fully containerised. | 🔒 Private repo |
| omniBox → Autonomous Goal Delegation | Assign any goal to an AI — it decomposes the task, spins up specialised agents, uses your tools, and delivers results. No workflows to define, no steps to configure. Multi-provider (Groq / Anthropic) with Tavily, Slack and GitHub integrations. | 🔗 Repo |
| Cortex-AI → Intelligent Personal Workspace | Full-stack workspace unifying tasks, habits, goals and learning behind an AI coach. Recurring-task engine, streak tracking, gamified XP/levels, calendar time-blocking and an AI auto-planner, on React + Vite + Zustand with an Express API and Firebase Auth. | 🔗 Repo |
| MCP Remote Server | A remote Model Context Protocol server implementation — exposes tools and resources over the wire so any MCP-compatible client can discover and call them. | 🔗 Repo |
| Triage Agent | Autonomous agent for automated triage: classifies incoming work, routes it, and acts on it without a human in the loop for the routine cases. | 🔗 Repo |
| Finance Tracker | Full-stack application to track, categorise and visualise personal finances. | 🔗 Repo |
| CodeCrafters Shell (Java) | A POSIX-style shell written from scratch in Java — tokenising, quoting, builtins, redirection and process execution. | 🔗 Repo |
> whoami
Ankit Pandey | Computer Science @ BITS Pilani
> pwd
~/work/ai-agents
> ls
Agents/
MCP/
Full-Stack/
Applied-ML/
DevOps/
> cat focus.txt
Multi-Agent Orchestration
Model Context Protocol
LLM Tooling & Evaluation
Full-Stack Product Engineering
> cat goals.txt
Ship agents that survive production.
Contribute to Open Source.
Build in public, measure everything.