I build the layer between AI agents and everything else — MCP gateways, agent orchestration, and production LLM systems.
Proof, not promises: at Mosaic Wellness I built and operate the org-wide agent platform — an MCP gateway federating hundreds of tool servers behind one endpoint with per-task capability routing, token budgets, and failure isolation. It cut organization-wide LLM token consumption ~93% (a 15× reduction) while becoming the default way agents access internal tools.
ctx-gateway — a clean-room, open-source MCP gateway: capability registry with embeddings-based retrieval, per-task tool-subgraph routing, hard token budgets, failure isolation, and a reproducible benchmark (tool-def tokens in context, task success, p95 latency). 14-day public build log.
| Project | What it is |
|---|---|
| AgentX | Personal AI assistant with memory, reasoning, and multimodal capabilities |
| mimicus | MCP-integrated agent system |
| WapiBot | Production WhatsApp booking assistant — FastAPI backend with an AI brain combining rule-based workflows and learned planning |
| VerifAI | JEPA-powered visual verification platform targeting Indian D2C return fraud |
| omen | Scene-aware JEPA render accelerator — research rendering engine |
| Fairdoc AI | AI-augmented medical triage & healthcare network |
| Compounding Engineering | Local-first AI engineering agent that learns from your code (DSPy) |
Biomedical R&D (3 yrs) → product management on India's first MCP-driven marketplace → production AI engineering. Every system I ship gets a gate that can fail and a number that can be read.
Reach: phant0mhriju@gmail.com · LinkedIn · Mumbai / remote



