You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
BARH is a Corrective Intelligence technology that I have been developing to work alongside AI systems and AI agents.
The idea came from a simple problem: an AI system can produce an answer or take an action, but the system itself may not have a separate mechanism to check what happened afterward, measure the result against a target, and correct it when needed.
BARH is designed to do this independently. It measures the current result, compares it with the intended target and boundaries, identifies the deviation, reduces the possible corrective actions, applies a controlled correction, and measures the result again before declaring it verified.
The BARH Core is built around six main parts
IFL — Input & Fact Layer
LBS — Logic & Boundary System
UAD — Unified Assessment & Deviation
DME — Deviation Measurement Engine
Reduction
Correction + Verification / READY
The core has already been implemented and tested. I have also usede it in practical experiments with real measurements and physical systems. These tests were useful not only when they worked, but also when they exposed problems in the verification logic that still need to be fixed and tested.
The next stage is to apply BARH directly to real AI systems and agentes.
I am also developing BARH Guard as an independent layer around AI systems. The Guard is intended to examine proposed outputs or actions and provide a separate control path for allowing, correcting, retrying, or escalating them.
The main question I want to test is whether this independent corrective layer can actually improve the reliability of AI systems. I do not want to assume that it works. I want to measure it through controlled experiments and compare systems with and without BARH.
The funding would allow me to continue development, build the AI integration and Guard layers, run these experiments, and bring in two specialized team members to support the technical work.
The main goal is to take the BARH Core that I have already built and tested and move it into real AI systems and agents.
I want to find out, through actual testing, whether BARH can make AI systems more reliable by checking their results, measuring deviations, trying controlled corrections, and verifying the result again.
To do this, I will:
* Develop the AI integration layer around the existing BARH Core.
* Develop BARH Guard to work independently around AI models and agents.
* Connect BARH to real AI systems without changing their underlying models.
* Test the full process from assessment and deviation measurement to Reduction, correction, re-measurement, and verification.
* Run the same tests with and without BARH so the results can be compared.
* Keep the results of both successful and unsuccessful tests and use them to improve the system.
If the funding is available, I also plan to bring in two specialized team members to help with the AI integration, testing, and verification work.
The goal is not to assume that BARH works. The goal is to build the next stage and te
st it properly.
The funding will be used to continue developing BARH and move it from the current tested Core to the next stage of development with real AI systems and agents.
Most of the work will be technical. I will use the funding to:
* Develop BARH Guard and the AI integration layer.
* Build test environments for AI models and agents.
* Run real experiments and compare systems with and without BARH.
* Improve the correction and verification parts based on the results.
* Build the tools needed to measure and document the experiments.
* Cover the development and testing costs of the project.
* Bring in two specialized team members to help with AI integration, testing, security, and verification.
The BARH Core itself is already built and tested. The purpose of the funding is to continue from this point, develop the missing parts, and test BARH properly with real AI systems.
I want the funding to go mainly into people, development, testing, and the equipment or services needed to run th
e experiments.
I am currently the main person working on BARH. I have been developing the architecture, writing and testing the core, and running the practical experiments myself.
BARH is not just an idea at this stage. I have built the BARH Core and tested its main components through software tests and practical experiments. I have also worked with real measurements, real sensors, physical intervention, and re-measurement to test the corrective loop.
Some of these experiments have worked as expected, while others exposed problems that still need to be fixed and tested. I consider this part of the development process and keep the results rather than presenting the system as already finished.
If the project is funded, I plan to add two specialized team members. Their roles will focus on areas such as AI integration, testing, security, and verification.
My current role is to lead the technical direction, architecture, development, and experiments, and to continue building BARH into a system that can be properly tested with real AI systems and
agents.
The main risks are technical. BARH will need to work with different AI systems and agents, and it may take time to find reliable ways to measure their outputs and verify corrections.
Another risk is that some of the correction methods may work in one type of AI system but not in another. The experiments may also show that some parts of the current BARH design need to be changed.
I plan to deal with these issues through staged development and repeated testing. I will measure the results at each stage and use the findings to improve the system rather than making assumptions about what should work.
If the project is funded, I also plan to bring in two specialized team members to help with the technical development, testing, and verification.
The intended outcome is a working and well-tested BARH system that can be evaluated with real AI systems. If some parts do not work as expected, the experiments will show us where the problems are and what needs to be improved before moving
further.
I have developed BARH independently and have funded the work myself so far.