I build web products and AI systems.

Problem

Support tickets contain useful answers, but turning them into a reliable knowledge base takes repeated review.

What I built

I built a seven-queue pipeline for ticket embedding, clustering, drafting, verification, publishing, RAG deflection, and gap detection. Socket.IO streams updates to the interface.

Results

• 46 integration tests reported passing.
• 0 SSN/card violations in a 784-record audit.
• 20% holdout set in the draft-and-verify workflow.

These figures describe the project’s reported tests and audit, not a guarantee for every input.

Stack

• React
• Node.js
• Express
• MongoDB Atlas
• Redis
• BullMQ
• Socket.IO
• OpenAI API

Problem

Using different AI providers can fragment a developer’s coding workflow.

What I built

I built a shared provider layer with streaming chat, prompt compression, token tracking, and conversation history inside VS Code.

Results

• 600+ marketplace downloads.
• 5 AI providers.
• Local inference through Ollama.

Download count reported in my resume; it is not a live usage counter.

Stack

• TypeScript
• VS Code Extension API
• esbuild
• LLM APIs
• Ollama

Problem

Sustainability reporting needs both document evidence and consistent calculations.

What I built

I built a document retrieval pipeline with Titan Embeddings v2 and FAISS. The language model extracts information; Python calculates GHG, water, and POSH metrics.

Results

• 8 BRSR categories evaluated.
• 1.0 context precision.
• 1.0 context recall.

Scores are from the documented local benchmark, not a claim of general accuracy.

Stack

• React
• Node.js
• Python
• FastAPI
• AWS Bedrock
• FAISS
• BullMQ
• AWS S3