Products I have built and led, and open-source projects I have contributed to. My role and the supporting sources are listed with each.
Built at KnackLabs · Open source
A runtime for AI employees. Teams define their responsibilities, control access to tools and data, approve actions and review activity.
I built Gantry at KnackLabs to operate agents through the channels teams already use, including Slack, Teams and Telegram. Its job is to make access, approvals and activity explicit.
Built at KnackLabs · Open source
The software factory our engineers use to build with coding agents. Agreed requirements, implementation, checks and human review stay connected through release.
I built Forge at KnackLabs to give agent-assisted development a defined process. The repository carries the agreed scope, decisions, checks and review evidence, so the team can inspect how a change reached release.
Built · Engineering skills and fresher hiring
An AI-taught bootcamp for freshers, from system design to debugging and production readiness. Students build projects and evidence that hiring teams can assess.
A coding agent teaches through practical work across six skills: system design, debugging, execution planning, AI-assisted engineering, production readiness and technical communication. Learner profiles record projects and reasoning. Interviews then verify what candidates built and understood.
Led at KnackLabs
AI code review informed by a knowledge graph of the codebase.
Led at KnackLabs
Voice agents for outbound campaigns, with analysis of recorded calls.
Initial implementation of the JavaScript MCP adapter. The LangChain repository credits @vrknetha and KnackLabs for that initial implementation; the project has been maintained by LangChain and its contributors since.
Initial implementation of the official Firecrawl MCP server, built at KnackLabs. Firecrawl's announcement credits Ravi Kiran for that initial implementation.
Three pull requests merged in January 2026: a host-only bash command for chat sessions, a Plivo provider for voice calls, and surfacing tool failures when an agent produces no output. Contributor, not maintainer.
Have a problem you want to work through?
Talk on LinkedIn