REAL PROJECTS

Five products we actually built

Not concepts: every project below has either a running demo or public source code. We spell out the problem, how we solved it, and the stack we used.

AI & Machine Learning

Floodiction

Water-level and flood-warning forecasts for 61 monitoring sites across Vietnam's 7 flood-prone regions.

Problem: Heavy rain and flash floods arrive fast, yet residents and local operators usually find out only once the water is already rising. Weather and water-level data sit in separate places, with no short-horizon forecast per location to act on.
Solution: The system pulls hourly weather data, engineers its own features, then runs two XGBoost models per region: a regression model forecasting water level in metres and a classifier forecasting warning level 0–3. Forecasts cover five horizons — 1h, 3h, 6h, 12h and 24h — surfaced on a map with charts and alerts.

Stack

N°02 · 2026Demo video

Healthcare Platform

A practice-management system for dental clinic networks, from booking through examination, payment and admin.

Problem: A multi-clinic dental network has to run booking, clinical records, treatment plans, prescriptions, payments and staff permissions — usually across disconnected tools that are hard to reconcile and leave no audit trail.
Solution: Four NestJS services on the Bun runtime behind a single gateway, each data domain owning its own PostgreSQL database. It covers the whole loop: booking by clinic, specialty or doctor → examination (diagnosis, treatment plan, e-prescription, orders) → payment and refunds → admin oversight (RBAC, audit logs, reports). Plus an OCR service for identity checks and a LangGraph booking assistant.

Stack

AI & Accessibility

Real-time ASL sign language recognition through a webcam, with an AI conversation assistant.

Problem: Deaf and hearing people lack an instant interpretation tool between them, and anyone learning sign language has nowhere to practise with immediate feedback on mistakes.
Solution: MediaPipe extracts 21 hand landmarks from each webcam frame, builds a 63-dimension feature vector, then classifies 36 signs (A–Z, 0–9) with a confidence score, tracking up to two hands at once. Around it sits a learning platform: topic articles, a Google Gemini assistant, and user/admin roles.

Stack

Blockchain & Traceability

Blockchain traceability for shrimp larvae, from hatchery to grow-out pond.

Problem: Shrimp larvae buyers have no way to verify a batch's hatchery, certificates or environmental conditions. Paper records are easy to alter, while export markets keep raising their traceability requirements.
Solution: Every larvae batch gets an identity on a Cosmos SDK chain: each supply-chain event is recorded immutably, documents live on IPFS, and IoT sensors stream environmental readings. A cryptographically signed QR code lets a buyer scan and read the full history, while decentralised identifiers (DIDs) and verifiable credentials authenticate each participant.

Stack

EdTech & E-learning

An online course and certificate management platform with a chatbot supporting learners.

Problem: A training provider needs one place to manage courses, lessons, certificates and payments across three different roles, while learners keep asking the same questions over and over.
Solution: A three-layer ASP.NET Core platform on SQL Server: instructors build courses, chapters and lessons, learners enrol and study, and certificates are issued automatically on completion. It adds admin course approval, payments, achievements, rankings, reporting, and a Gemini chatbot answering learners.

Stack

Ask us about this project