Kazi Hassan A.

I build AI systems that do real work: agents, automations, MCP servers, internal tools and generative video. Then I keep them running in production. In 2026 I co-founded Trivo, an AI-native video studio. We’ve delivered 6,000+ AI-assisted videos to clients in all 50 US states, growing from 300+ a month to 800+. I built most of the tooling behind that with AI. WHAT I’VE BUILT THIS YEAR • AI video app (prototype): an order ID goes in and a finished 30-second 1080p video comes out. It has a 9-stage pipeline, AI vision, FFmpeg rendering, 25 automated QC checks and human approval before anything publishes. It has 2,100+ automated tests and was verified read-only against a live dev API. I built it in 2 days by directing multi-agent Claude Code. • 3 custom MCP servers / Claude plugins that connect Claude to a company’s admin API, Google Sheets and Frame io. They default to dev, need an explicit flag to touch production, and read secrets at runtime. • Browser-agent automation that replaced a manual delivery workflow (Sheets → Frame io → admin portal). The first full run handled 5 of 5 orders with zero failures. Later runs flagged broken links instead of failing silently. • A Claude Code agent with approval gates, spending caps and resume support. It also checks that every photo in an order shows the same property. On one test order, that check halted the run when an order’s 30 photos showed 11 different homes. • A data pipeline across ~5,000 orders, with parallel agents and rate-limit retries, feeding a live analytics map. • A production website: TypeScript on Cloudflare Workers, a serverless contact form, CI on every push and data-verification tests. • Reusable Claude/Codex skills, for example one that measures a reference video’s editing style (42.8 cuts a minute, 1.37 s average shot) so it can be reproduced. • AI virtual staging that keeps furniture consistent across camera angles, using per-job scene memory. • ML R&D: local depth-model 2.5D parallax with measured QC, and frame-by-frame hallucination checks on AI-generated clips. TOOLS Claude (Code, Cowork, skills, MCP) · OpenAI Codex · Kling · Seedance · Higgsfield · ElevenLabs · Suno · Depth Anything · Python · TypeScript · FFmpeg · OpenCV · Google Apps Script · Docker · Cloudflare HOW I WORK • AI provides the speed, and a human checks everything that ships. Every system I build has review gates, budget caps and a dev mode before it touches production. • I measure. When I say something works, a test or a run log shows it. • I learn fast. In 2026 I went from browser agents (May) to my own MCP servers (June), a multi-agent full-stack app (September) and local ML models (October). GOOD FITS • Automating a workflow your team repeats every day, across Sheets, Drive, CRMs, admin portals and APIs • Connecting Claude or ChatGPT to your systems through MCP servers, skills or agents • Internal AI tools and prototypes, from idea to working app • Generative video and image pipelines at volume • An R&D partner who tests new models the week they launch Native Bengali, fluent English. I work async with clients all across the globe. Tell me the task your team repeats every day. I’ll tell you how I’d automate it and what it would take.