Ubik Studio — Portfolio | Blaise Albis-Burdige

The Problem

Existing AI assistants hallucinate citations and treat "research" as question-answering. Ubik Studio was built for the researchers who can't afford that — people writing theses, long-form journalism, policy analysis, or any work where citations actually get checked.

The product sat at the intersection of LLM tooling and epistemics: every claim the AI made had to trace back to a real source the user had in hand.

What I Built

I designed and built the entire product from scratch as the sole technical co-founder.

Full-stack application: Next.js frontend, Python backend
Multi-model routing — researchers pick their provider (OpenAI, Anthropic, etc.) and switch mid-workflow without losing context
PDF annotation engine — upload a paper, highlight a passage, and the AI cites from that passage specifically rather than paraphrasing from memory
Citation verification pipeline — cross-references AI-generated claims against uploaded sources to flag potential hallucinations
Local-first architecture — institutions with data privacy requirements can run Ubik on-premise

Status

Ubik wound down in 2026. The core idea — that every AI-generated claim must trace back to a real source the user holds — carries forward into my work on rrxiv, an open protocol for verifiable, agent-native research.