Glassbox — The AI Detective Lab

Hero
Proof

Inspiration

AI can give answers in seconds—but most of the time, we do not know why it reached them. For research, investigations, verification, and everyday decision-making, a confident answer without visible reasoning is not enough.

That is the problem we wanted to solve with Glassbox — The AI Detective Lab: make AI investigation feel less like a black box and more like a transparent detective desk.

Our idea was simple: give users a place to bring messy information, investigate it with AI, connect the important clues, and receive conclusions that remain understandable and traceable.

What it does

Glassbox helps users investigate a question from evidence rather than blindly trusting a final answer.

A user can provide notes, documents, links, statements, or raw text and ask an investigative question. Glassbox then helps them:

Extract important people, places, dates, claims, and events
Identify relationships and patterns between pieces of evidence
Surface contradictions, missing context, and unanswered questions
Build a clear evidence trail behind each conclusion
Separate what is supported by evidence from what is only an AI-generated hypothesis

The result is an AI workspace that feels like a detective lab: clues come in, connections become visible, and every conclusion can be inspected.

How we built it

We designed Glassbox around one core principle: the answer should never be more visible than the evidence behind it.

The product flow begins with an investigation board, where users add information and define the question they want to answer. Our AI layer structures the input into useful investigative signals, then organizes them into a traceable view of facts, connections, confidence levels, and open questions.

Instead of treating AI output as final truth, Glassbox presents it as a working case file. Users can review the supporting evidence, challenge assumptions, and keep refining the investigation.

We focused heavily on making the experience approachable. The interface uses the familiar visual language of case files, evidence cards, connections, and timelines—so users can understand complex information without needing to understand the underlying AI system.

Challenges we ran into

The hardest challenge was balancing intelligence with honesty.

It is easy for an AI product to sound certain. It is much harder to clearly show uncertainty, distinguish evidence from inference, and avoid making unsupported claims look authoritative.

We addressed this by designing Glassbox to make uncertainty visible. Every insight should have context: what evidence supports it, what assumptions were made, and what still needs to be verified.

Another challenge was turning an investigation workflow into something that feels simple instead of overwhelming. We iterated on the experience to keep the user focused on one question at a time while still preserving the full chain of clues behind the answer.

What we learned

Building Glassbox taught us that transparency is not a secondary feature for AI—it is part of the product itself.

People do not only want faster answers. They want to know:

What information was used?
Which claims are actually supported?
What is uncertain?
What should be checked next?

We learned that trustworthy AI should help users think better, not merely produce more text.

What's next for Glassbox

We want to evolve Glassbox into a complete investigation workspace for researchers, journalists, analysts, students, compliance teams, and curious people who need more than a confident answer.

Our long-term vision is simple: