productivity-boost.com
DE – Deutsch
Menu

Florian Standhartingera one-person AI lab in Passau

Software that decides — and tells you how sure it is.

Route a ticket, approve an action, score a draft: one API call, one probability per option. I build the models, the API and the open network that serves them.

Get an API key (opens in a new tab)Read the quickstart (opens in a new tab)

State “I need to update my invoice address.” · business account

Question Which support queue fits the request? s1-fast

  1. billing0.87
  2. technical0.09
  3. account0.04
Show the JSON

Request · POST /v1/systemone

{
  "model": "s1-fast",
  "state": {
    "message": "I need to update my invoice address.",
    "account_type": "business"
  },
  "questions": {
    "route": {
      "type": "choice",
      "instructions": "Choose the support queue that best fits the request.",
      "criteria": {
        "billing": null,
        "technical": null,
        "account": null
      }
    }
  }
}

Response

{
  "id": "dec_route",
  "model": "s1-fast",
  "answers": {
    "route": {
      "type": "choice",
      "choice": "billing",
      "confidence": 0.87,
      "probabilities": {
        "billing": 0.87,
        "technical": 0.09,
        "account": 0.04
      }
    }
  },
  "usage": {
    "input_tokens": 351,
    "output_tokens": 0,
    "decisions": 1
  },
  "tier": "eu"
}

State action: delete_workspace · role: viewer

Question Is this action allowed under the role policy? s1-fast

  1. no0.96
  2. yes0.04
Show the JSON

Request · POST /v1/systemone

{
  "model": "s1-fast",
  "state": {
    "action": "delete_workspace",
    "actor_role": "viewer"
  },
  "questions": {
    "allowed": {
      "type": "noul",
      "instructions": "Is this action allowed under the stated role policy?"
    }
  }
}

“noul” is the probability of yes.

Response

{
  "id": "dec_guard",
  "model": "s1-fast",
  "answers": {
    "allowed": {
      "type": "noul",
      "noul": 0.04
    }
  },
  "usage": {
    "input_tokens": 298,
    "output_tokens": 0,
    "decisions": 1
  },
  "tier": "eu"
}

State A short release-note draft · rubric: clarity and factual support

Question Rate the draft against the rubric. s1-pro

  1. high0.62
  2. medium0.33
  3. low0.05
Show the JSON

Request · POST /v1/systemone

{
  "model": "s1-pro",
  "state": {
    "draft": "Short release note draft",
    "rubric": "Clarity and factual support"
  },
  "questions": {
    "quality": {
      "type": "score",
      "instructions": "Rate the draft against the rubric.",
      "criteria": [
        "low",
        "medium",
        "high"
      ]
    }
  }
}

Response

{
  "id": "dec_score",
  "model": "s1-pro",
  "answers": {
    "quality": {
      "type": "score",
      "score": 1.57,
      "confidence": 0.71,
      "legend": {
        "0": "low",
        "1": "medium",
        "2": "high"
      },
      "probabilities": {
        "0": 0.05,
        "1": 0.33,
        "2": 0.62
      }
    }
  },
  "usage": {
    "input_tokens": 512,
    "output_tokens": 0,
    "decisions": 1
  },
  "tier": "eu"
}
Example responses from the Decision Models quickstart. decisionmodels.io/docs/quickstart (opens in a new tab)

How it fits together

  1. Measure

    JevBench

    My open benchmark ranks 194 open-weights decision models on accuracy, calibration, latency and cost.1 (note)

    benchmarkheaven.com (opens in a new tab)

  2. Serve

    Decision Models

    The strongest open models behind one API, with an EU tier.

    decisionmodels.io (opens in a new tab)

  3. Decentralise

    REAX

    Independent GPUs serve the same models; every answer is checked.

    reax.co (opens in a new tab)

01 — Products

What I’m building now.

Decision Modelsdecisionmodels.io

One API, many open decision models.

Decision models don’t write text. They read a situation and a set of options and return one answer per option with a probability, in a single forward pass. That makes them fast and cheap enough to sit in the middle of any workflow. Decision Models hosts strong open ones behind one API that speaks the same request format as Jev, TypeSafe AI’s decision model.2 (note)

decisionmodels.io (opens in a new tab)Quickstart (opens in a new tab)Models (opens in a new tab)

The decisionmodels.io home page: headline “Inference for open-weights decision models.” next to a request and response sample.The decisionmodels.io home page: headline “Inference for open-weights decision models.” next to a request and response sample.
Screenshot of decisionmodels.io, October 2026.
Models
s1-fast · s1-pro · s1-vision (image + text) · quyet-1-large
Interfaces
REST API, hosted MCP server, Jev-compatible requests
EU tier
Processed only in the EU, payloads not stored, never used for training

REAXreax.co · Bittensor subnet 92

An open network that serves decision models.

On REAX, anyone with a capable GPU can host the models, validators check every answer, and the network rewards correct, fast serving. It runs as subnet 92 on Bittensor and has been live on mainnet since 14 October 2026. Decision Models is the customer-facing product on top.

reax.co (opens in a new tab)Mine with your GPU (opens in a new tab)REAX Engine on GitHub (opens in a new tab)

The reax.co home page: headline “AI by the people, for the people.” and the serve, verify, reward, improve loop.
Screenshot of reax.co, October 2026.
Miner pools
Four: s1-fast and s1-pro (the API models) on NVIDIA GPUs, plus s1-lite for small GPUs incl. Jetson Orin and s1-cpu for plain CPUs
Competitions
An engine track for the fastest correct serving and a generalist decision-model track, both open since October 2026
Onboarding
Give your coding agent one link; it practises on a local chain with test TAO first, and you approve every payment

02 — Built in public@airesearch12 (opens in a new tab)

Built in public.

I share most of what I build on X as @airesearch12. A few of those posts travelled further than I expected.3 (note)

More from the lab

Everything else on GitHub (opens in a new tab)

03 — Earlier work, 2014–2021

Before decision models.

VR and AR apps, game-engine tools, profilers and business systems. The images are from my old site.

  • Flat illustration of a dashboard with pie chart, bar chart and line graph above a world map with three location pins, on a teal background

    Business systems

    BMS & AREA

    Business monitoring and geo-information tools for a DAX-listed insurer: millions of records in real time, including contract structures and natural-catastrophe risk. Lead developer and architect.

  • Room Designer VR logo above a virtual living room with a sofa, a furniture selection menu and a VR controller menu

    VR · Steam, 2017

    Room Designer VR

    Plan a room while standing inside it: draw the floor plan on the ground, then place furniture, doors and windows. For HTC Vive.

  • Close-up of smart glasses at a football match; the lens shows a red record dot and the text "Tap to save last 10 seconds as video ..."

    AR glasses · Vuzix

    Rewind Recorder

    The camera keeps a short buffer; one tap saves the last ten seconds as a video. For the moments you would otherwise miss.

  • Unity editor screenshot with a candelabra mesh and its collider, and the Non Convex Mesh Collider component with numbered setup steps

    Unity Asset Store, 2017

    Non Convex Mesh Collider

    Collision detection for concave meshes, a gap in Unity’s physics at the time.

  • Collage of profiler screenshots showing call trees, a timeline, a pie chart and source code with timings

    .NET tools

    Insight Profiler

    A profiler that goes beyond performance and shows how a .NET process actually works.

04 — About

Portrait of Florian Standhartinger

Hi, I’m Florian.

I wrote my first program in 1987, when I was six, and never really stopped. productivity-boost.com has been my company since 2014: VR and AR apps, developer tools, business systems, and now mostly AI, from decision models to agents and benchmarks.

I like building the whole thing, from model and engine to API and website, and putting it out early. Next to the lab, I work as a Lead IT Consultant at msg systems ag.

05 — Contact

Building with decision models, or want to?

Need the EU tier, volume or a custom model, or just want to talk shop? Write to me. I read everything and reply to almost all of it.

info@productivity-boost.com

Or start right away