I build AI features in production — and the observability to prove they work.
Senior full-stack developer, 7 years. Lately: multi-stage LLM pipelines with retrieval, prompt design, and observability built from scratch across the back end. This page is the argument: no framework, no web fonts, no JS bundle. It loaded before you finished reading this sentence.
this page · measured
lighthouse (lab) · lcp field
0perf
0a11y
0seo
0slcp · field
how this site works
- deploy
- Cloudflare Pages + one Worker reading vitals from KV. ~$0, sub-1s globally.
- stack
- Hand-written HTML/CSS, system font stack. No framework, no build, no JS bundle.
- vitals
- The three scores are Lighthouse (lab), enforced in CI. The lcp is real field data — a rolling p75, web-vitals → Worker → KV.
- at work
- FlowLinker: tracing, metrics, DB insights and log-trace correlation — all as code.
what i'm building now
FlowLinker is a sales-conversation intelligence platform. I built the multi-stage LLM pipeline that turns call transcripts into structured requirements and solution-fit signals — retrieval, prompt design and LLM observability — plus the backend observability the whole system is judged by. Read the architecture →
selected work
- FlowLinker—AI sales-call platform: production LLM pipelines, prompt design, observability from scratch.
- Reppa · MonGym—Full-stack on a multi-service fitness product: React Native/Expo + Java/Quarkus on GCP.
- Empego—Scaled Angular to 30k MAU; cut build 45s → 10s; bundle −50%.
- AlleyCorp Nord—Resolve + Reside Health: React/Next dashboards, Python + Kafka, Stripe; design system −40% dev time; RN app 4.8★.
- GoTo—Messaging + meeting features for millions of users; WebRTC, RxJS, WebSockets.
- Earlier—Finance, ERP and BI: 1C ERP, Qlik/Power BI over 100M+ records.