Supercore · AI systems engineering
Practical AI systems for teams with real data, real users, and real constraints.
I help turn AI ideas into working software: knowledge systems, assistants, voice interfaces, and the infrastructure needed to run them without drama.
What I help with
Useful AI, wired into the systems people already depend on.
Turn messy knowledge into usable AI
RAG systems for documents, websites, support notes, legal material, and internal knowledge bases.
pgvector, hybrid search, citation grounding, reranking
Build assistants that can take action
Chat and voice assistants that call tools, remember context, and work inside real business workflows.
function calling, MCP, workflow gates, audit trails
Move prototypes into production
The less glamorous work that makes AI useful: auth, tenant isolation, streaming, logs, costs, and failure handling.
Next.js, FastAPI, Postgres, Vercel, Fly.io, local GPUs
Run AI on your own infrastructure
Self-hosted LLM inference for teams that can't send data to a third-party API — on-prem or private cloud, sized around compliance, cost at scale, and latency.
vLLM, quantization, GPU sizing, private RAG
Private AI deploymentsProof of work
A few systems that show the range.
Catalyst AI
Flagship runtimeA shared assistant platform with RAG, voice, tool calling, tenant-scoped data, and admin controls.
Used as the base runtime behind multiple products and the site assistant.
π.Law
Legal AIA legal case-management and document-search system designed around sensitive client data.
Uses a proxy boundary, Postgres search, and Catalyst for AI reasoning over legal material.
Forensic AI Studio
Investigation toolsA private evidence-analysis workspace for searching documents, mapping entities, and analyzing audio.
Built under real legal constraints with document ingestion, vector search, and agent tools.
Silicon Smackdown
Voice AIA real-time AI talk-show experiment with multiple voice personalities and live conversation flow.
Demonstrates low-latency audio, WebSocket orchestration, and Gemini Live API work.
How I work
Less theatre. More working software.
I am useful when the problem has messy inputs, sensitive data, awkward integrations, or unclear production constraints. The goal is not to add AI everywhere. The goal is to make one important workflow faster, safer, or easier to operate.
Good fits
Bring a workflow, dataset, or product idea where generic chatbot wrappers are not enough.
- Legal, evidence, support, or internal knowledge workflows
- Assistants that need tools, memory, or integrations
- Voice or real-time interfaces with latency constraints
- AI systems that need privacy, observability, or cost control
Contact
Tell me what you are trying to build.
A useful first message includes the workflow, the data involved, who will use it, and what would make the project worth doing.