Building a customer support platform for a European manufacturer. Hybrid RAG, streaming replies, multi-tenant infrastructure. 100K+ messages in production.
AI platforms, Chrome extensions, full-stack web apps, and cloud infrastructure. One or two engagements at a time.
Start a projectProduction RAG systems, streaming replies, and multi-tenant infrastructure. Chat4Team processes 100K+ messages with 80% reply acceptance.
Full-stack TypeScript in React, Next.js, and Astro. Type-safe end to end, with real auth, real payments, and clean migrations.
Manifest V3 at scale. 55,000+ users, 26 storage migrations, zero data loss.
.NET and Azure Functions at TCS-scale. Self-managed Podman and Caddy on bare VPS. Zero-downtime deploys with health checks and rollback paths.
Customer support platform for a European manufacturer. Hybrid RAG, streaming replies, multi-tenant infrastructure. 100K+ messages in production.
New-tab productivity dashboard re-architected for Manifest V3. 55,000 daily users, 26 storage migrations, zero data loss.
Multi-agent research and outreach pipeline. Async job queue with Graphile Worker, type-safe end to end with Zod.
Compliance-scanning controls in C#/.NET and Azure Functions. Deployed across hundreds of thousands of Azure subscriptions at TCS.
Every engagement starts with the problem, not the framework. Time invested upfront on scope and constraints prevents months of rework later.
Small, reversible pull requests. CI/CD, health checks, and migrations built in from day one — never retrofitted at the end.
I build, deploy, and support the systems I design. Handoffs are documented, predictable, and engineered to be handed off.
Architecture decisions, trade-offs, and postmortems are documented. The next engineer should be able to read the system without talking to me first.
Customer support platform for a European manufacturer. Hybrid RAG over documents, scraped pages, and resolved support tickets. Streaming replies, human-in-the-loop operator tools, zero-downtime deployments on self-managed infrastructure.
Two years in production. The architecture here — hybrid retrieval with a reranker, streaming responses, and operator tools — has become the template for every AI system I build.
Long-form writing at r44j.dev on the systems I've built — RAG pipelines, Chrome extension migrations, zero-downtime deployments. Documentation of what worked, what failed, and the decisions behind each call.
One or two engagements at a time. AI platforms, Chrome extensions, full-stack web applications, and cloud infrastructure. Scoped on the first call.
