grantmaking.ai Launch Round
With the increasing use of large-language models in legal compliance, procurement, and governance processes, there are issues with trusting their outputs because the models may generate confident outputs with no reliable evidence and without disclosing uncertainty. The use of AI to assist in making high-impact decisions creates risks for organizations.The ConsensoOS solution uses a retrieval-augmented generation (RAG) architecture combined with specialized analysis pipelines for contracts and policies to recommend decisions based on an evidence-based approach. The recommendations provided by ConsensoOS are linked to source documents (via citations, confidence rating, and audit trail) and do not rely on opaque generative responses. ConsensoOS also has guardrails in place to prevent hallucinations; it will automatically map compliance with common regulatory frameworks, including GDPR, ISO 27001, SOC 2, HIPAA, and the EU AI act; and include an interactive "playbook" feature to allow users to interrogate documents with fidelity to their source.
The initiative seeks to further develop these features through creating an open evaluation framework in order to effectively measure citation fidelity/hallucination rates/uncertainty consistent across repeated evaluations/reasoning grounded use in enterprise AI.Deliverables from this effort will include:
1.Open evaluation benchmarks
2.An AI safety evaluation toolkit
3.Enterprise document grounding framework
4.Open documentation and deployment guides
5.Research findings about how AI can be used in legal and compliance not just trustworthy.
- Complete development of the open evaluation framework
- Build benchmark datasets
- Expand retrieval and citation infrastructure
- Improve anti-hallucination evaluation
- Cloud infrastructure and hosting
- Independent testing
- Documentation and open-source release