About Boutique · Bayern · Solo

One specialist. One verdict. One responsibility.

Verdict is independent diligence run by one person. One person builds the pipeline. One person operates it. One person reads your brief and signs the verdict. Straight out of Bayern.


It started as a system for myself. The diligence was a byproduct.

The honest version: Verdict was not the plan. I built an AI system for my own work, and the pipeline behind it turned out to do something I had not set out to sell.

  1. the side quest Month 0

    A personal AI system, built for me

    I wanted a tool that actually understood my work and remembered it. So I built Lucid, a personal AI system, piece by piece. Not a product. A thing I needed.

  2. the turn The architecture

    It became a memory graph and a research pipeline

    Lucid grew a persistent memory graph, semantic vectors, and a multi-agent research pipeline that could fan out across sources, weigh them, and pull a question apart instead of summarising it.

  3. the moment The realisation

    The same pipeline reads a company cold

    I pointed that pipeline at a company instead of a research question, and the output looked like investor-grade diligence. Sourced claims, weighed signals, an honest verdict where the evidence actually pointed. That was the moment Verdict made sense.

  4. today 9 months in

    Verdict is that pipeline, productized

    Then I read what passes for AI research elsewhere. Most of it is a plausible summary with no sourcing and no discipline. Verdict exists so the diligence you buy is not one more of those. It is built to the standard I held my own system to.

Most AI "research" is a confident guess.

A fluent summary, no sourcing, no discipline, no separation between what is known and what is asserted. That is exactly why so much AI output reads well and means nothing when a real decision is on the line.

When Verdict reads a company for you, it is not at the quality you see everywhere else. It is at the quality I built for my own system.

Solo. Bayern. Engineer, not a fund.

Fabian Ilg. Nine months building and running a self-hosted AI system, a multi-agent research pipeline I wrote myself, and an honest opinion on what holds up under scrutiny and what does not.

Fabian Ilg, Lucid AI Labs
Fabian Ilg · Lucid AI Labs
240,049 Lines shipped
9 mo Solo, in production
~8,000 Memory nodes
24/7 Own GPU

Direct. No layers, no stake in the deal.

Independent means no analyst pool passing your brief along and no fund whose answer is decided before the work starts. You talk to the person who runs the pipeline and checks the highest-stakes facts by hand.

What you do not get
  • An analyst pool you never meet
  • An account manager between you and the work
  • A fund pushing the deal it is invested in
  • A conflict of interest baked into the answer
What you get
  • The person who built the pipeline reads your brief
  • No fund affiliation, no stake in whether you proceed
  • The highest-stakes facts checked by hand
  • One name on the report, accountable for the call

The multi-agent pipeline Verdict runs on.

If you want to see how deep the engineering goes, look at the bench it runs on. The same memory layer, sourcing discipline, and self-hosted inference that carry my own system carry straight into every report.

verdict:~ engineering bench runs every report
Multi-agent research pipeline, agents fan out, weigh sources, then reconcile
~8,000 memory nodes as 768d vectors, retrieved by relevance
Self-hosted inference on a dedicated GPU, EU only, no data resold
240,049 lines across the system, nine months, one engineer
See the method See the report

Where I share what I build.

If you want to see who is actually behind your report, the code is public. No marketing repo, the real thing.


Have a company you're evaluating? Request a report.

Read by me personally. One specialist, one independent read, no fund behind it and no deal to push. If the fit is right, I go deep. If a signal is thin, the report says so.

Questions: Fabi@lucid-ai.app