# Yrikka

> Yrikka builds vision AI that adapts to the mission. ORION is a vision-AI lifecycle platform: it generates training data, tests vision models and VLMs to find the operational context where they fail, and fine-tunes them for new targets, sensors and environments. ATLAS is a mission simulation platform: it generates mission worlds from a sentence, a map location or a target photo, so teams can test a full autonomy stack before field trials. ATLAS can be integrated with simulation environment that the user prefers including Isaac, Unity, and Gazebo. APEX is the API for running Yrikka evaluations from your own pipeline.

Yrikka serves autonomy builders, defense primes and government teams. Its research and development is funded through programs with the U.S. Army (SBIR, including Phase II work), the U.S. Air Force (SBIR, and an STTR with Vanderbilt University) and USSOCOM. ATLAS is awardable on the CDAO Tradewinds Solutions Marketplace.

Key facts:

- Yrikka's key products are ORION, ATLAS and APEX.
- ORION covers three steps. Data: generate and prepare labeled data, including RGB, infrared and SAR. Test: evaluate vision models and VLMs across mission-relevant operational conditions and adversarial inputs. Adapt: agentic fine-tuning of the customer's own or an open-weight model to improve the model performance in mission conditions.
- ATLAS generates physics-based aerial and maritime mission worlds with real terrain and structures, controllable weather, light and visibility, and synchronized sensor views with ground truth. Customers connect their own autonomy stack through an API without sharing source code.
- ATLAS specifications: domain adaptation in under 20 minutes; under 20 reference images; visible (EO) and thermal (IR) sensors, with radar (SAR) in beta and RF in alpha; ground truth exported as COCO, Pascal VOC or a custom schema; runs in AWS, AWS GovCloud, or air-gapped on edge devices.
- To talk to the team or book a technical demo, use https://www.yrikka.com/work-with-us

## Products

- [ORION](https://www.yrikka.com/orion): Vision AI lifecycle platform. One worked example follows a truck detector from one clear-day image through generated fog, night, rain and snow, a failure test, fine-tuning and a successful retest.
- [ATLAS](https://www.yrikka.com/atlas): Mission simulation for autonomous systems. Covers world generation from text, map coordinates or a target photo, condition changes, synchronized render, simulation and depth views, and specifications.
- [APEX API](https://www.yrikka.com/api): Developer quickstart for running condition-based model evaluations from your own pipeline, with per-condition precision, recall and F1.

## Use cases

- [Sensor adaptation](https://www.yrikka.com/sensor-adaptation): RGB-to-infrared and RGB-to-SAR generation from limited aligned data, with downstream detection results.
- [Infrared video for counter-UAS](https://www.yrikka.com/infrared-video): Six generated counter-UAS scenes with aligned RGB and thermal video, with detector results that show where a model fails.
- [Use case overview](https://www.yrikka.com/#use-cases): Summary of all use cases across ORION, ATLAS and APEX.

## Research

- [Research](https://www.yrikka.com/research): Yrikka's papers and technical writing, with venues, awards and links to code and datasets.
- [Text-to-Model technical report](https://www.yrikka.com/text-to-model): How synthetic data, automated evaluation and agentic training produce a vision model, with a detection case study in heavy snow (YOLOv11-Large on the DAWN benchmark).
- [CausalARC: Abstract Reasoning with Causal World Models](https://arxiv.org/abs/2509.03636): Peer-reviewed; NeurIPS 2025 LAW Workshop spotlight paper; presented at the Amazon AGI Trusted AI Symposium 2026. Dataset on Hugging Face with more than 1,000 downloads: https://huggingface.co/datasets/jmaasch/causal_arc
- [Inference-Time Scaling of Diffusion Models for Infrared Data Generation](https://arxiv.org/abs/2511.07362): Peer-reviewed; NeurIPS 2025 Learning to Sense (L2S) Workshop.

## Company

- [About Yrikka](https://www.yrikka.com/about): Mission, research and government program relationships.
- [Work with us](https://www.yrikka.com/work-with-us): Contact form for demos and use-case discussions.

## Optional

- [Photorealism in AI: FLUX.1 Krea vs Dev](https://medium.com/@maxim.clouser/photorealism-in-ai-comparing-the-ai-look-of-flux-1-krea-vs-dev-829c13f1a1ed): Technical article comparing real and generated images in the frequency domain.
- [APEX quickstart on GitHub](https://github.com/YRIKKA/apex-quickstart): Example code for the APEX API.
- [ATLAS on Tradewinds](https://www.tradewindai.com/tsm-awardable?search=Yrikka): ATLAS listing on the CDAO Tradewinds Solutions Marketplace.
- [Website privacy](https://www.yrikka.com/privacy): How website inquiries are handled.