ADP-013: Development with CoopHive of a Multi-hop RAG System for Reproductive Longevity

Summary

This proposal seeks approval to allocate two milestone allocations of $25K each (paid in vested ATH) from the DAO Treasury to enter into a fund a partnership under the form of a services agreement to be executed with CoopHive.

AthenaDAO and CoopHive will build together an advanced multi-hop Retrieval-Augmented Generation (RAG) system tailored to AthenaDAO’s research in reproductive longevity, with an emphasis on ovarian health.

The system will transform AthenaDAO’s scientific corpus into a domain-specific RAG plus fine-tuned LLM system that supports multi-hop reasoning, transparent citations, and reliable scientific answers.

Funds will be disbursed in ATH equivalent, vested monthly over 12 months, and tied to milestone progress.

Motivation

AthenaDAO is dedicated to advancing research in female reproductive longevity. However, the scientific literature is fragmented across thousands of studies, making synthesis and discovery challenging.

By entering into this agreement with with CoopHive, AthenaDAO can:

Transform its curated corpus into an interactive research knowledge system.

Enable multi-hop scientific reasoning with reliable citations.

Fine-tune open-source AI models specifically for reproductive longevity research.

Expand the impact and accessibility of AthenaDAO-funded research for scientists and the DAO community.

This agreement aligns directly with AthenaDAO’s mission by building infrastructure that accelerates scientific discovery in women’s health.

Details

Phase 1: RAG Development & Evaluation - $25,000

• Embedding & Integration ($7,000): Process AthenaDAO-provided papers into 4–6 vector databases with varied embeddings, chunking, and preprocessing strategies. Enable querying across all.

• Human Evaluation Setup ($5,000): Develop evaluation rubrics for accuracy, reasoning depth, and completeness. Test with 20 curated QA pairs.

• Cross-Database & Automated Evaluation ($6,000): Build an LLM-driven evaluation pipeline; compare human vs. AI evaluation on ~30 open questions.

• Large-Scale Validation & Reporting ($7,000): Validate with 100 expert QA pairs + 300 LLM-generated questions. Deliver a comprehensive evaluation report.

• Publishing & Dissemination: Cover costs for paper submission and conference travel.

Phase 2: RLHF Optimization & Deployment — $25,000

• Fine-tune lightweight open-source LLMs using RLAIF, LoRA, cross-encoders, and LM-as-a-judge.

• Train models on AthenaDAO’s curated datasets for domain-specific accuracy.

• Deploy public endpoints for AthenaDAO and CoopHive frontends to query databases in real time.

Deliverables:

• Fine-tuned, open-source models specialized in reproductive longevity.

• Transparent evaluation reports.

• Usable query endpoints for researchers and the AthenaDAO community.

Funding

• Total Request: $50,000 (paid in ATH equivalent at time of disbursement).

• Source: AthenaDAO Treasury.

• Payment Structure:

  • $25,000 for Phase 1 (RAG Development & Evaluation).

  • $25,000 for Phase 2 (RLHF Optimization & Deployment).

• Vesting: Equal monthly payments over 12 months.

• Accountability: CoopHive will provide quarterly progress reports; AthenaDAO retains the ability to pause vesting if milestones are not met.