Joined as a founding engineer, led the development of the tooling infrastructure, improved the developer experience internally and externally, and ran multiple growth experiments during the company’s early search for product-market fit.
- Rewrote the early tooling infrastructure, delivering more robust infrastructure and an 11× throughput improvement
- Designed and developed core tooling infrastructure from the ground up, including a scalable tool-execution runtime, a versioning service, and an OpenAPI-to-tool generation pipeline
- Built reusable abstractions and internal developer tooling that accelerated implementation while reducing regressions and operational risk
- Built Composio SDK, SWEKIT (scored 48.6%, second place in SWE Bench at the time), and the plugin ecosystem as part of early growth experimentation
Joined as one of the founding engineers, led the development of the core stack and multiple open source frameworks to improve the developer experience.
- Reduced deployment complexity on the user platform by automating environment configurations
- Improved the user experience by developing a client for the user platform and reducing deployment time by 95–99%
- Designed and developed the POC for a no-code/low-code platform for developing autonomous services
- Coordinated release cycles across the core open-source frameworks
- Built an orchestration framework that streamlined the development, testing, and deployment of services built using the agent frameworks
- Introduced custom static-analysis tooling that accelerated development and reduced production issues
- Built plugins enabling on-chain interactions across multiple blockchain networks and wallets
- Built internal tooling that reduced service deployment time by up to 80%
- Designed and built IPFS-based package-management infrastructure and an extensible plugin ecosystem around it
- Developed autonomous-service prototypes using the core stack to validate its capabilities and potential use cases
Accepted as a summer intern in GSoC 2021, working on a low-code platform for developing neural networks with TensorFlow at Emory School of Medicine.
- Reduced workflow complexity by introducing a simple and coherent development process
- Reduced development time by improving the no-code UI design and implementing a JSON-to-Python transpiler for converting graph-like data structures to boilerplate Python code
- Introduced an in-browser code editor to reduce iteration time