Solo-built a cross-platform Bible study app in Flutter (iOS, Android, macOS, Windows, Linux) on an offline-first architecture, with an AI study assistant, original-language word study, and in-app subscriptions.

Achievements / Tasks

• Designed an offline-first data layer: everything reads and writes locally, and Supabase Postgres syncs per-row with last-write-wins, so the app stays fully usable with no connection.
• Built an AI study assistant with retrieval over the user's own notes and journals plus a public-domain confessions corpus (Postgres full-text + pgvector), streaming replies over SSE. Per-tier model routing and prompt caching keep unit cost under the subscription price.
• Built original-language word study: Strong's alignment maps any translation's words back to Greek/Hebrew lexicon entries, with occurrence counts and rendering distribution per word.
• Wrote Python content pipelines against Supabase Storage to ingest public-domain commentaries, confessions, and letters, and to generate Kokoro TTS audio with per-verse timing for offline listening.
• Built a synced hymn score player (Verovio → SVG with a custom cursor renderer).
• Shipped the commercial layer: RevenueCat IAP subscriptions with webhook-side entitlement grants, GitHub Actions release pipelines to TestFlight and Play internal testing, and Universal/App Links deep linking.

Tech Stack

Languages: Dart, Python, PostgreSQL

AI / ML: pgvector, Kokoro

Infrastructure: Supabase, RevenueCat, GitHub Actions

Libraries: Verovio

Frameworks: Flutter

Architected a self-hosted AI voice platform on dual NVIDIA DGX Sparks, replacing third-party services and eliminating per-minute API costs.

Achievements / Tasks

• Built real-time voice pipeline with faster-whisper (STT), vLLM + Qwen3 / GLM-4.5 (LLM), and Kokoro (TTS), achieving fast end-to-end latency.
• Designed a node graph FSM for call routing covering intent, order collection, account lookup, returns, and escalation using structured exit tokens.
• Vector DB for RAG, building isolated per-project knowledge bases with LangChain ingestion pipelines for multi-format data.
• Self-hosted LiveKit (Go / Node.js / Rust) as WebRTC orchestration layer and built a Svelte 5 frontend dashboard for agent monitoring and configuration.

Tech Stack used at Employer

Languages: Python, Typescript, Go, Rust

AI / ML: faster-whisper, vLLM, Qwen3, GLM-4.5, Kokoro, LangChain

Infrastructure: NVIDIA DGX Spark, LiveKit

Frameworks: Svelte 5

Work closely with stakeholders and upper management in healthcare. Initially hired as a front end web developer but given opportunities to work on back end .NET frameworks (4, 6, 8) and database schemas in Microsoft SQL Server. Depending on the project, including previous ones, I switch between front end or back end frameworks.

Favorite project: Build and maintain functional Learning Management Software ( https://orientation-plus.org ) coded in Blazor (.Net Core), Typescript, and MS-SQL with Dapper ORM. Facility and/or Healthcare Staff can use GUI UI tools, with animation, to create Orientation courses for nurses. This also includes early stages allowing designers to create mobile friendly courses.

Tech Stack used at Employer

Languages: C#, Typescript, SQL

Libraries: Reveal JS, Interact JS, DragSelect, DapperORM

Frameworks: .NET (4, 6, 8), Blazor WebAssembly, Svelte

Developing a CRM applicatiton in Real Estate and match making system to automate the less need for real estate agents.

Tech Stack used at Employer

Languages: Typescript, PostgreSQL

Tools: Supabase Firebase

Frameworks: Svelte Kit