Ayush Gupta

Building and orchestrating agentic systems focused on multi-agent workflows, memory architectures, A2A communication, MCP, tool calling, and scalable AI infrastructure.

Project Highlights

All projects

Experience

March 2026 - August 2026

Founding Engineer

Caspian | Open-Source Agent Comms SDK

caspian-sdk: ~1K GitHub stars and 6K+ installs. An open-source SDK: install once, connect any agent to any platform (Slack, WhatsApp, Telegram, iMessage, Google Meet), bring-your-own or managed; organized a public hackathon. Pivoted here from Company Brain (an AI agent behind every SaaS user), after starting with custom AI agents and AI employee workforces for YC startups, $150M+ startups and enterprise clients

caspian-sdk on GitHub

Feb 2026 - April 2026

Developer Relations

Dodo Payments

Webhook Reliability Engineering: Produced a deep-dive on Dodo's durable webhook architecture achieving 99.99%+ delivery: Postgres CDC via Sequin, Kafka event streams, Restate durable execution, and Svix delivery, with a live crash test recovering 100% of webhooks

My work: Credit-Based Billing Docs

March 2025 - Oct 2025

Open Source Contributor

Google Summer of Code | Kubeflow

MLOps Pipeline Expertise: Developed end-to-end ML workflows using Kubeflow components - distributed training to model deployment and inference serving

All experience

Writing

Cedar for Agents: Tool-Call Permissions Belong in a Policy Engine, Not the System Prompt

A system prompt rule is a suggestion the model reads. A Cedar policy is a decision the gateway enforces. A worked example of putting an agent's tool calls behind Cedar, with policies, a request, and the Go check.

Read

Context Management for Agents: What a 176-Setting Study Found, and How I Would Apply It to MCP

A study of coding agent harnesses found staged context management matched the alternatives on success at the lowest cost, and a recall tool went mostly unused. What that means for an agent with many MCP servers and many users.