Award W3villa Technologies

Rising Star
The Rising Star award came after an intense, rewarding stretch at W3villa Technologies, where I took AI out of experiments and into products that enterprise customers use every day.
- 120K+
- agent interactions a month
- 98%
- intent accuracy
- 38%
- lower LLM spend
- 60%
- fewer prompt incidents
The work behind it
- AI agents customers actually use. Led multi-tenant agent platforms for HRMS, CRM and LMS products — past 120,000 interactions a month, at 98% intent accuracy and under two seconds a response.
- Reliable tool calling. Fine-tuned small models with LoRA and QLoRA so they call tools predictably — 34% more deterministic, with 41% fewer hallucinations.
- Search that feels instant. Rebuilt retrieval around hybrid keyword + vector search with reranking, taking latency from 420ms to 273ms and making a million enterprise records easier to find.
- Smaller bills, faster answers. Moved serving to vLLM and TensorRT with quantisation and semantic caching — three times the throughput and 38% lower overall LLM spend.
- Voice that keeps up. Built real-time voice assistants that hold a conversation in under 300 milliseconds.
- Guardrails for a team. Introduced evaluation and a version-controlled prompt system, cutting prompt-related incidents by 60%.


