I sit with numbers, find what they're actually saying, and build systems that make organisations act on it autonomously, at scale. Over 10 years across BFSI, CPG, energy, and healthcare; MBA from Liverpool Business School, UK.
The 10-day process that becomes 2 hours. The 3% error rate that drops to 0.5%. The best AI systems are the ones nobody notices they just work.
I fix real bugs in production-grade AI infrastructure — agentic frameworks, LLM platforms, vector databases, AI tooling. The kind of issues that silently corrupt data, break under concurrency, or surface only at scale.
Merged
Notable open PRs
| Repo | Contribution |
|---|---|
| livekit/sip | In-dialog request routing via proxy Record-Route; media-drain before BYE to prevent voicemail clipping |
| livekit/livekit | Room re-create race: pre-clear room state so same-name recreate gets a fresh SID |
| livekit/protocol | SIP header value cap raised to 4KB for carrier tokens (X-Twilio-CallToken) |
| livekit/agents | Soniox final-span timing: all timed tokens, never regress end_time |
| openai/openai-python | Drain remaining bytes after [DONE] before closing streaming responses |
| langgenius/dify | PG logical replication fix; agent file type filtering; MCP client timeout fallback |
| chroma-core/chroma | Embedding function correctness, FTS robustness, dependency hygiene |
| crewAIInc/crewAI | Async native tool calling in the invocation loop |
| mem0ai/mem0 | Log redaction for sensitive content; embedding dims auto-propagation |
| huggingface/smolagents | Code agent reliability and tool-use edge cases |
Open to architecture challenges, enterprise AI strategy, freelance engagements, or startup ideas in the GenAI space. If you're building something that matters, let's talk.
Data first. Always.