A distributed system across GKE and AWS Lambda that uses fine-tuned LLMs to generate compliance, risk, and fraud insights from streaming transactional data. gRPC + Connect RPC + Kafka + GraphQL across services in Python, Go, and TypeScript.
Scalable distributed backend (Go + PostgreSQL + Redis cache), TypeScript/Next.js/Redux frontend with Clerk auth and Stripe payments, and an Apache Kafka pipeline for real-time analytics and observability. Open-source.
Built on Firebase Genkit and LangChain. Built-in chat history, response caching, query expansion, endpoint auth, and Retrieval-Augmented Generation. CI/CD-published to NPM as `@oconva/qvikchat`.
Built on QvikChat. Provides architecture and data strategy for low-latency intent recognition without training supervised NLU models. Cloud Firestore-backed; published to NPM as `@oconva/intento`.
Cross-platform app with expiration/consumption tracking, RAG-powered AI assistant, voice command integration, and personalized waste-reduction insights. Built around UN Sustainable Development Goals 2, 9, 12, and 13.
University of Ottawa capstone. Architected the full AI component: ingestion from public university sources, embedding generation, vector storage (Pinecone, Mongo), and a custom LangChain executor pipeline with multi-model support (GPT-4, PaLM 2, Llama 2 Chat).
University of Ottawa software-engineering project. Built around a custom Dispatcher-Action-Handler state pattern that decoupled the data layer from Firebase, plus Firestore-driven real-time updates between clients and chefs.
A platform that connects buyers with non-profits selling kids' art. React + Bootstrap.
A Go pipeline that runs RANSAC iterations concurrently to find dominant planes in a point cloud. Exercise in Go concurrency primitives.
Framework for LLM-powered applications. Documentation contributions.
Google's language-neutral data-serialization format. Documentation updates.
JS/TS snippet extension. Added support for async arrow function snippets.