A production-readiness platform for software built with AI coding tools. It catches the gap between "it works" and "it's ready for users": exposed secrets, missing authentication, unsafe database access, missing rate limits, unoptimized network calls, memory leaks, N+1 queries, and other performance and reliability issues AI-generated code tends to miss.

I built the scanning engine, caplora-cli, using static analysis rather than asking an LLM to guess, every finding backed by evidence. For issues it can safely fix, it writes the patch, opens a pull request, and re-runs the check to confirm it's actually resolved, not just flagged. The web app connects to a GitHub repo, runs the scan, and gives a single readiness call: blocked, ready with risks, or ready.

A classroom management mini-app API built to pull homework distribution out of Facebook groups and Messenger chats, where it's currently scattered alongside unrelated and unmoderated social content, and into a dedicated, safe channel inside the chat apps families already use. Teachers post homework into class group chats, students submit work without leaving the conversation, and parents get automatic updates, all without a separate app or login.

It's built as a mini-app API, not a MongolNet-specific product: a mapping-table pattern lets the same backend plug into any chat platform's group-messaging interface, not just MongolNet, so schools aren't locked into one super-app. I built the full stack: an Expo/React Native frontend, a Fastify and Drizzle ORM backend on PostgreSQL, and Redis for caching and background jobs, with a CI/CD pipeline running 29 unit tests and 11 end-to-end tests before every deploy. It's deployed on AWS with Nginx, PM2, and SSL, and handles MongolNet's real authentication flow, including session-based auth exchange, role mapping between Mongolian terms and internal roles, and real-time updates over WebSocket.

A peer-to-peer currency exchange platform I co-founded and lead the technical side of, built to eliminate the scam risk that comes with informal currency exchange between individuals.

I designed the schema, built the backend in Python and PostgreSQL, containerised it with Docker, and deployed it via CI/CD. I also wrote the technical specifications that translate business requirements into buildable features, then review each release against the spec to catch mismatches before it ships. As Co-Founder and CTO, this is where most of my hands-on engineering judgment and my instincts for pitching a product's story come from.

A full-stack booking platform I built for the Mongolian market. It gives appointment-based businesses (barbershops, salons, spas, karaoke rooms, restaurants) their own custom online booking page where customers can book 24/7 without calling or messaging.

It covers the full loop: a real-time customer booking flow, a business owner dashboard with a weekly calendar, a super admin panel to manage every client business from one place, two-way Google Calendar sync so bookings land on the owner's phone automatically, a QR code and Instagram card generator per business, automated SMS reminders through Twilio, and a fully Mongolian-language UI throughout. It runs as a managed SaaS.

A family memory platform I built to solve a problem I've had since I was a kid: family history mostly lives in a few people's heads and disappears when they do. Family members add memories in whatever format is easiest, photos, voice recordings that transcribe themselves, written stories, or documents, into one shared, private archive.

The core feature is a retrieval augmented chat interface: ask the archive a question in plain language and it answers using only the family's own stored memories, citing the specific one behind every claim. The part I'm proudest of is gap detection, which actively scans what's been recorded, spots what's missing, and prompts the right family member to fill it in, backed by an automatic family tree and an interactive timeline. Built on Next.js and TypeScript, with Supabase for the database, storage, and auth, pgvector for retrieval, and OpenAI for the grounded answers, transcription, and embeddings. Every family's data is isolated with row level security.