Project NANDA Fellowship Program
Fast Builder AI Fellow
High-ownership builder role embedded in the NANDA team — prototype, ship, and maintain infrastructure for the Internet of AI Agents
Position overview
Project NANDA (Networked Agents and Decentralized AI) is an initiative led by Prof. Ramesh Raskar, building the foundational infrastructure for an Internet of AI Agents.
The core problem: the existing internet was designed for humans. DNS resolves domain names typed by people. HTTP serves documents to browsers. PKI authenticates websites for users clicking padlocks. None of it handles billions of autonomous agents that need to discover each other, verify capabilities, and transact in milliseconds — without a human in the loop.
NANDA's architecture addresses this across four layers:
Discovery: a federated NANDA Index — DNS for agents — that resolves newly spawned agents sub-second and routes queries to the right registry without a central bottleneck.
Identity: AgentFacts, cryptographically signed and schema-validated JSON-LD documents that describe what an agent can do, who operates it, and how to connect securely.
Federation: a quilt-like index design, currently replicated across 15+ universities and partner institutions, where any organization can host its own registry while remaining globally discoverable.
Interoperability: native support for Anthropic's MCP, Google's A2A, and Microsoft's NLWeb, so agents built on any major framework can participate from day one.
NEST (NANDA Sandbox and Testbed) is the platform where agents are deployed, registered, and orchestrated on this stack. It's the primary thing you'll be building on and contributing to.
This is not a traditional research position. It's a high-ownership builder role embedded in the NANDA team. You'll prototype, ship, and maintain infrastructure — some of it ends up in papers, some in live deployments, some on partner events.
You'll face open-ended problems with no ticket, no spec, and no obvious answer. You'll make architectural calls on the fly. You'll context-switch between wiring up a hackathon demo and thinking carefully about how the AgentFacts schema should handle capability versioning. The role demands both speed and substance — and the judgment to know which one the moment calls for.
What you'll do
- NEST platform: deployment tooling, agent orchestration, registry APIs, and the developer experience on top of the NANDA stack
- Agent registration and discovery flows — implementing and stress-testing the NANDA Index against real multi-agent workloads
- Protocol integration: MCP servers and clients, A2A agent cards, AgentFacts schema validation and tooling
- Full-stack infrastructure: backends, dashboards, simulation harnesses, and anything else the system needs
- Live demos and showcase systems for NANDA events and hackathons — things that actually run in front of an audience
- Research-adjacent prototypes: when a paper describes a system, someone has to build it. That's you.
- Exploring what comes after current protocols — federated trust, privacy-preserving discovery, multi-hop agent routing
Expertise you may have
What We're Looking For
- No YoE floor. No YoE ceiling. Ten years in and want to get into agentic AI properly? Great. Two years in and already building agents? Also great. We care about what you can do, not how long you've been doing it.
- Technical depth is the one non-negotiable. Everything else is learnable on the job.
- You have strong CS fundamentals — you understand how systems talk to each other, where things break under load, and why certain architectural decisions compound over time
- You can read an unfamiliar codebase and get oriented fast
- You're hands-on across the stack: you can write a backend service, build a frontend to demo it, containerize it, and debug it when it falls over in production
- You vibe code — you use AI tools to move fast without losing structural integrity
- You can own a problem end-to-end: scoping, building, testing, shipping, presenting
- You don't wait to be told what to do next
Bonus (Not Required)
- You know MCP or A2A at a practical level, or you've worked with agent frameworks (LangGraph, AutoGen, CrewAI, or similar) and can get up to speed fast
- You understand how agentic systems actually work — not just "LLM calls a tool" but how agent graphs, tool invocation, context management, and multi-agent handoffs compose in practice
Communication & collaboration
- High visibility — your work gets presented at various events and feeds into published research
- Real ownership — no hand-holding, no micromanagement, no artificial constraints
- Fast feedback loops — you'll know if something works within days, not quarters
- You'll be working at the frontier of agentic AI infrastructure, not catching up to it
Hiring process
This role is fast and sometimes chaotic. Priorities shift. Problems are ambiguous. Timelines are tight. If you need detailed specs, structured onboarding, and clear swim lanes — this probably isn't the right fit.
Apply now