CASE 01 — CODENAME: BAZAAR
Under NDA
Screens withheld at client's request
AI Agents Marketplace
ClientConfidential
ScopeFull platform — discovery to deployment
Under the hoodMulti-agent orchestration · LLM routing · sandboxed execution
StatusShipped end-to-end
The build story
A full two-sided marketplace — agent publishing, discovery, sandboxed execution, usage metering, and an orchestration layer routing requests across multiple LLM providers without them stepping on each other. The kind of platform teams usually staff ten people for; we architected, designed, and shipped it lean, end-to-end. That's all we can say — and honestly, the silence is the best part of the review.
CASE 02 — CAUSALITH · OUR OWN
Causalith
WhatAI research copilot — your research buddy
Under the hoodRAG · Neo4j knowledge graphs · vector embeddings
DoesPaper ingestion → citation-grounded, graph-linked answers
The build story
Our own product — and our proof of craft. Papers are parsed, chunked, and embedded into a vector database; entities and claims are lifted into a Neo4j knowledge graph; every answer is retrieval-grounded and cited back to the exact passage. Ingestion pipeline, graph construction, RAG layer, and the interface you see here — all built in-house. When we say "production AI," this is what we mean.
CASE 03 — CODENAME: LORESMITH
Under NDA
Game + client name confidential
In-Game AI Companion
ClientConfidential — game studio
ScopeCustom local LLM with a designed personality
Under the hoodFine-tuned on lore · runs local · Discord-native
StatusLive, in character 24/7
The build story
A local model fine-tuned on the game's entire lore corpus, wrapped in a personality designed to never break character. Persistent memory, moderation guardrails, and native Discord integration — running on the studio's own hardware, so no API bills and no player data leaving the building. Players talk to it like it's part of the world, because it is.
