Work

What we’ve built

Real systems, shipped. Private on-prem AI, retrieval platforms, and production apps — here’s a look at how we work and what it delivers.

Local AIAgentic Offices

Private AI for a public affairs firm

For a Washington, DC public affairs firm

The problem

The firm wanted AI for research and drafting — but its work involves confidential client strategy, and staff pasting that material into public chatbots was a genuine security risk.

What we built

  • A private AI system running on the firm’s own hardware — prompts and documents never leave the network
  • A fleet of research agents that monitor news and policy sources, compile briefings, and draft first-pass research on a schedule
  • Access controls and full audit logging, managed by the firm’s internal IT

The result

  • →Research briefings that once took hours are turned around in minutes
  • →Zero client data sent to any outside AI provider
  • →A repeatable pattern the firm’s IT team fully controls

Stack

On-prem open-weight LLMLocal inference (Ollama)Scheduled multi-agent pipelineAccess control & audit log

Proof that serious AI can run entirely in-house for firms that can’t send data to the cloud.

RAG SystemsTailored Applications

InfoMuck — regulatory intelligence with RAG

For compliance teams tracking regulatory change

The problem

Compliance teams track regulatory change across thousands of sources. General chatbots guess, cite the open web, and can’t tell you what actually changed for a specific client.

What we built

  • Grounded chat: every answer comes from an indexed knowledge graph of 4,200 curated sources — with citations staff can verify — never the open web
  • Client Vault: a living dossier per client, updated by event-driven crawls and weekly digests
  • Desks: proactive monitoring that flags relevant changes before anyone asks
  • Briefing pipeline: a multi-agent system that turns findings into finished slide decks
  • Multi-tenant from day one, so each firm’s data stays isolated

The result

  • →A working platform you can try live today
  • →Answers that compliance staff can trust, because every claim is cited back to a source

Stack

Postgres + pgvectorLlamaIndexClaude & other LLM APIsGoogle Cloud
Visit InfoMuck

Proof we can connect AI to a large, messy body of documents and make the answers trustworthy.

Tailored ApplicationsLocal AI

NPC Talk — an AI messaging app on iOS & Android

A shipped consumer app on both app stores

The problem

Build a social messaging app where AI characters hold real, ongoing conversations — without paying frontier-model prices on every single message.

What we built

  • A Flutter app shipped to both the App Store and Google Play from a single codebase
  • A custom backend where AI characters stay resident and keep their personalities across conversations
  • Characters running on local open-weight models, so per-message inference cost stays near zero
  • An open API and MCP server, so outside AI assistants like Claude can join conversations inside the app

The result

  • →In production on both the App Store and Google Play
  • →AI running affordably in production — near-zero cost per message

Stack

FlutterNode / Python backendOllama (open-weight)MCPGoogle Cloud
App Store Google Play

Proof we can take a mobile app through both app-store reviews and run AI affordably in production.

Have something like this in mind?

Tell us what you’re trying to build — we’ll tell you honestly if we can help.

Book a callSee our services