About

I'm Sriram Ramakrishnan, a product manager working at the intersection of stablecoins and AI. I build open tools that make AI agent payments safe before they're signed.

Why this blog exists

AI agents can now pay for things on their own, over protocols like x402, ACP, AP2 and MPP. On stablecoin rails a confirmed payment is final: there's no chargeback if the agent gets it wrong. So the checks have to happen before the agent signs. That's what I write about here, with runnable code wherever I can.

What I've built

• SRA-RiskGate-4B: an open 4B model that reads a payment request and returns approve, hold or reject with reason codes. It runs locally via GGUF and Ollama.
• SRA Stablecoin Risk Bench: the benchmark and training data behind it.
• sra-riskgate on PyPI: the Python SDK, with integrations for MCP, LangChain, CrewAI and ElizaOS.
• Live demo: try to trick it into approving a bad payment.
• Paper: SRA-RiskGate: A Benchmark and a 4B Model for Gating Stablecoin Payments Made by AI Agents.

Also writing on

• DEV.to
• Hugging Face

If you're building agents that move money and want to compare notes, leave a comment on any post. I read them all.