I'm Gopika, a CS student at IIT Mandi who builds agents, the crypto-native kind. AI that doesn't just answer questions, but reads the onchain world, reasons about it, and acts inside it.
What pulls me isn't models, it's agency: giving AI somewhere real to operate. A protocol. A social graph. An ecosystem with money and governance moving through it in real time. Most AI is a clever interface bolted onto the internet. I'm more interested in what happens when an agent is actually native to it, and what it can be trusted to do once it's there.
That obsession shows up in what I ship. Sentinel, a governance and trust layer for a workforce of AI workers that spend real money, took 2nd at the Base × Privy hackathon. Credence, an earned-authority passport for agents, came out of ETHGlobal New York. Echo is autonomous ecosystem intelligence for Base, live and watching governance, grants, and onchain activity so it surfaces what matters before it's obvious. Alter Ego reads how people behave on Farcaster and predicts whether two builders would actually collaborate. dep-graph rethinks how agents plan: tools emit semantic entities that become the prerequisites for the next move, so an agent can reason about what it has versus what it still needs.
The crypto half is just as deliberate. I was selected into the Uniswap Hook Incubator to build at the protocol layer (concentrated liquidity, v4 hooks, dynamic fee models) because you can't design agents for DeFi without understanding how the rails actually move. Solidity, mainnet state, tick math: I'd rather read the contract than the docs.
By day I'm a SWE intern at MSB Docs, where I shipped an AI document-understanding pipeline that turns messy scanned forms into fillable documents at ~85% field recall. Not crypto, but the exact instinct behind the onchain work: take something chaotic and unstructured and turn it into data a machine can actually act on. Underneath all of it is genuine systems work, async pipelines, deterministic scoring, onchain state, caching that holds under load, but the engineering is the means, not the identity. The point is agency that survives contact with production while nobody's watching.
Right now I'm deep in v4 hooks, agent trust and reputation primitives, and the x402 world where agents pay each other, the things that have to exist before autonomous systems can be handed anything that matters. It all comes back to a few stubborn questions:
what can an agent actually be trusted to do?
can coordination be automated without losing the humans in it?
what does AI look like when it's native to crypto, not bolted on?
When I'm not chasing those, I'm building the human side of the same problem: mentoring students, running workshops through DevRel Uni and SheFi, and dragging more people into technical spaces that used to feel locked. Coordination is coordination, whether the nodes are agents or people.
I take the work seriously. I've just never believed serious people have to be boring about it.
// status: caffeinated, mid-build, probably teaching an agent something it'll misuse.