$ whoami
VP Engineering @ KnackLabs (part of CAW Studios). Hyderabad.
I build AI agent infrastructure. Identity protocols. Cross-platform messaging. MCP contributor. Rust when it matters. TypeScript when it doesn't. LangChain Ambassador on the side.
I run an AI assistant from a spare MacBook that manages my LinkedIn, reviews my code, screens CVs, finds leads, and occasionally roasts my bad decisions.
$ cat /proc/current
Clawdentity — cryptographic identity + secure messaging for AI agents. Per-agent DIDs, Ed25519 signing, WebSocket relay, cross-platform webhooks. 12K lines of Rust. 70 tests. IETF Internet-Draft published. Works on OpenClaw, PicoClaw, NanoBot, NanoClaw. One binary. Zero deps.
upsk.to — an engineering bootcamp with no lectures and no videos. Your coding agent is the instructor — it teaches through conversation, remembers how you learn, and scores on evidence. Rust CLI + Cloudflare Workers behind it, pure conversation in front. Works with Claude Code, Codex, Cursor. Paste one prompt to start.
$ ps aux | grep knacklabs
| PID | Process | What it's running |
|---|---|---|
| 1 | symphony-forge | The assembly line for agent-built software. Claude Code plans, Codex executes, deterministic gates enforce sign-off → plan → build → ship. Adversarial "grill" review before every handoff. |
| 2 | gantry | Self-hosted runtime for production AI agents. Sandboxed workers, capability grants, memory that survives model swaps. Slack / Teams / Telegram / Discord. |
$ ls ~/projects --sort=stars
| Project | What | Stars |
|---|---|---|
| openclaw | Personal AI assistant runtime. Shipped Plivo voice calls, /bash command, tool-failure surfacing. |
 |
| firecrawl-mcp-server | Web scraping for AI assistants. Led initial implementation. |  |
| langchainjs-mcp-adapters | MCP ↔ LangChain bridge. Contributor. |  |
| clawdentity | Agent identity protocol + Rust CLI. |  |
$ cat /etc/opinions
- Small fine-tuned models on tool calling > generic large models for most workflows
- Nobody talks about the interface problem. Your agent is smart. Your users don't know what to ask.
- Everyone demos. Nobody ships. It breaks at the boring stuff — config, auth, retries, edge cases.
- Data quality > model size. Always.
- CLI agents are general-purpose. Not "coding tools."
$ uname -a
$ history | tail -5
- Launched upsk.to — bootcamp where the coding agent is the instructor
- Shipped Gantry — self-hosted agent runtime, prepping open-source release
- Built Symphony Forge — the harness KnackLabs ships every project through
- Published Clawdentity IETF Internet-Draft (RFC for agent identity)
- Running 15+ automated cron jobs from an AI assistant on WhatsApp