Products I have architected and shipped, from clinical AI and RAG systems to IoT, games and graphics experiments.
Learning platform with assessments, DB-driven product checkout, a high-traffic waiting queue, group-tier promo codes, bulk entitlements from CSV with an audit trail, and an admin studio for publishing course content. The launch generated about $700K USD in total sales in under 48 hours.
Pilot platform for structured caregiver programs: digital onboarding, cohort and session management, support plans, daily logs, passwordless sign-in, and coach and admin tooling.
Configurable, content-specific chatbot platform for clients, distributed as an npm SDK and CLI with a React chat widget. Retrieval is namespaced per tenant, with reranking, model routing, a semantic cache and input and output security guards. Kindred is the reference implementation.
Mobile app and platform that turns travel photos into ranked, structured micro-guides. Expo and React Native client, NestJS API, Supabase Postgres with private media, background image processing, and three automated deploy pipelines.
Personal Android app that teaches piano from zero using a connected keyboard. It reads every note over USB-MIDI in real time and runs a tutor that does not advance on a wrong note, with falling-note and practice modes, bar navigation and section looping, and per-bar mastery tracking that restarts near the mistake. A harmony analyzer names keys, Roman numerals and chords, and a structured lesson path covers reading, scales, intervals and chord progressions. It imports MIDI and MusicXML, plays back through a built-in synth, and works fully offline. In active development; the core has 28 unit-test files.
Specialized medical LLM for ambient listening, transcription, and structured chart generation. HIPAA-conscious RAG architecture; integrates with EHR systems and voice-based intake.
Production RAG chatbot powering this portfolio. It retrieves relevant site context from pgvector and prompts an LLM to provide grounded answers, with CORS-safe and cost-aware delivery.
Graph-based learner persona detection to personalize lesson paths. Neo4j + KNN; Flask inference API; optimized Cypher queries for latency and scale.
Schema-driven form builder used by financial and legal teams, with GraphQL and REST data flows, role-based permissions, and autosave. Optimized queries and CDN caching for global users.
Android IoT app to monitor and control high-pressure aeroponics. Telemetry and commands run over MQTT/HTTP, supported by backend dashboards, sensor charts, and ML-based growth recommendations.
End-to-end Sudoku solver: detects a grid from camera frames, recognizes digits, and solves the puzzle via backtracking. Built a dataset and augmentation pipeline, with model selection in Keras/TensorFlow.
WebGL-based procedural terrain using Gaussian noise + mesh generation; interactive camera + shading pipeline; focused on geometry + GPU rendering fundamentals.