AI Solutions Architect, Forward Deployed Engineer, Agent Engineer, Business Analyst, Full Stack Developer, Digital Marketing Specialist
I built a document pipeline for an Australian state government client. 90% F1 on one named benchmark, with every value linked to its source page.
An Australian state government client · via Silvatron Pty Ltd
The whole run: a deterministic parse, twelve LangGraph nodes, then two human review gates before the CRM.
Every tool here ships in a published case study.
Document pipeline · 7 tools · Read the case
Three projects, each with its status and how its numbers were measured.
Status: ProductionBuilt at Silvatron Pty Ltd
Claude Code, n8n, Bitbucket, Jira and Discord run first-pass reviews and diagnose failed pipelines. A person approves every fix.
Evidence: About 20–40 developer hours a month (estimate)
Stack: Claude · n8n · Node.js · Bitbucket
Status: Built, not launchedBuilt at CoralShades
A policy retrieval system where Postgres row-level security decides what each role can see, and every answer cites the document it came from.
Evidence: Role isolation enforced by the database, not the app; answers cite their source
Stack: React · TypeScript · Supabase · PostgreSQL
Status: Shipped
A phone-first scoring and live leaderboard app for a Sri Lankan New Year community event on 18 April 2026, built in two days.
Evidence: 237 commits over 17 and 18 April 2026, the last on event day.
Stack: Next.js · Supabase · TanStack Query · Tailwind CSS
Tell me how the work runs today and where it breaks. I'll tell you whether an agent belongs in it, and what a first version could look like.