HEO

Hypothesis-to-Experiment Orchestrator

Comment

Inspiration

Modern scientific research is fundamentally bottlenecked. Despite billions invested in lab automation and AI, critical steps remain painfully manual, siloed behind paywalls, and notoriously difficult to reproduce. This isn't just an academic inconvenience; it's a global equity issue.

Meet Alex, a public health researcher with an NGO. For Alex, every experiment is a battle: sifting through mountains of literature, replicating brittle methodologies from PDFs, and praying results aren't compromised by human error. This friction means critical hypotheses take months to validate, delaying life-saving interventions for the communities they serve.

We built HEO to fix this. We were inspired to create a future of open, reproducible science, accessible to everyone, everywhere, by providing a universally accessible, affordable, and trustless pipeline for turning ideas into validated experiments.

What It Does

HEO orchestrates the entire journey from idea to trustless, on-chain proof:

  • 🧠 Generate Hypotheses: Vertex AI’s Gemini and a MongoDB Atlas Vector Search RAG system ensure each research idea is grounded in real domain context.
  • 🖱️ Design Protocols Visually: Our new drag-and-drop Protocol Designer removes scripting headaches. Experiment design is now a simple click-and-connect process for any researcher.
  • 🤖 Execute in the Cloud: HEO can instantly triggers protocols on real cloud labs (Strateos/ECL/….) via our serverless orchestrator.
  • 🛡️ Validate with Zero-Knowledge: Circom + SnarkJS generate a zkSNARK proof that guarantees experiment integrity without exposing sensitive raw data.
  • 🔗 Anchor on Solana: The proof's hash is immutably written on-chain to the Solana mainnet, creating an irrefutable record.
  • 📦 Package for Open Science: All results and metadata are packaged in a FAIR JSON-LD format and pinned to IPFS.
  • 🏆 Reward Trust: A pilot token layer is being designed to reward validators and researchers for contributing to reproducibility.
  • 🔌 Plug & Play for Labs: We provide a ready-made GitLab CI/CD template for labs and developers to plug reproducibility-as-code directly into their existing workflows.

How We Built It

  • Frontend: Next.js App Router with a React-based drag-and-drop Protocol Designer featuring real-time schema checks.
  • AI & RAG: Vertex AI (Gemini) for hypothesis generation, with a hybrid RAG architecture powered by MongoDB Atlas Vector Search for providing relevant context.
  • Serverless Orchestration: Google Cloud Functions + Cloud Run handle all asynchronous tasks, with BigQuery logging for a complete audit trail.
  • Lab Automation: Direct REST API integration with Strateos/ECL/….
  • Proof Layer: Circom circuits and a SnarkJS prover for generating zkSNARKs.
  • Blockchain: The Solana network (@solana/web3.js) is used for anchoring immutable proof hashes.
  • Data Storage: An IPFS HTTP client pins all FAIR-packaged JSON-LD data.
  • DevOps: A GitLab CI/CD template was created to enable "reproducibility-as-code."

Challenges We Ran Into

  • Building an Intuitive UI: Creating a drag-and-drop interface that could handle complex, real-world lab schemas and provide useful error handling was a major UX and engineering challenge.
  • Optimizing zkSNARK Performance: Generating proofs for large experiment payloads was initially slow. We solved this by implementing an incremental proving strategy, which significantly reduced computation time.
  • Managing Schema Drift: Keeping our data representations consistent across JSON-LD, MongoDB, and on-chain formats required careful schema management and validation layers.
  • Designing Fair Tokenomics: It was a significant challenge to design a fair tokenomics model that properly aligns incentives for reproducibility without creating unnecessary on-chain bloat or exploitable loopholes.

Accomplishments that we're proud of

  • ✅ Achieved a true one-click "Hypothesis → Lab → Proof → On-chain" pipeline.
  • ✅ Deployed a zkSNARK proof anchored live on the Solana mainnet, which is reproducible by anyone and verifiable by a smart contract.
  • ✅ Created a zero-trust FAIR JSON-LD package for our results and pinned it to IPFS, making open data a reality.
  • ✅ Built a novel drag-and-drop Protocol Designer that empowers non-coders to design complex experiments.
  • ✅ Created a GitLab CI/CD template so any lab can plug reproducibility into their repo in minutes.
  • ✅ Designed a pilot token incentive layer to create a sustainable ecosystem around trustless reproducibility.

What We Learned

  • How to productionize Vertex AI for real-time scientific hypothesis generation with tight latency constraints.
  • The power of hybrid RAG + Vector search architectures with MongoDB Atlas for deep contextual understanding.
  • Best practices for zkSNARK circuit design and proving for WebAssembly environments.
  • The nuances of FAIR data standards, JSON-LD packaging, and pinning data at scale on IPFS.
  • The critical balance between blockchain immutability for proofs and flexible off-chain storage for large data payloads.
  • How providing CI/CD templates can serve as a zero-friction entry point for labs to adopt reproducibility-as-code.

What's next for HEO

  1. Launch Free Academic Tier: Roll out our open incentive pilot so labs and students can earn tokens for reproducible experiments.
  2. Expand the Protocol Designer: Add advanced templates, real-time monitoring, and collaborative design features.
  3. Roll Out CI/CD Templates: Create more templates to cover various hardware setups and lab providers.
  4. Launch the Hypothesis Marketplace MVP: An on-chain repository where verified experiments can be bought, forked, and reused.
  5. Publish Tokenomics Contract: Deploy our fair tokenomics model to the Solana testnet to gather community feedback.

Built With

  • bigquery-database-&-rag:-mongodb-atlas-vector-search-lab-automation:-strateos/ecl-rest-api-zksnark:-circom
  • blockchain:
  • ci/cd
  • client
  • cloud-functions
  • devops:
  • gitlab
  • http
  • ipfs
  • languages/frameworks:-node.js
  • next.js
  • react-cloud-&-ai:-google-cloud-run
  • snarkjs
  • solana
  • solana/web3.js)
  • storage:
  • typescript
  • vertex-ai-(gemini)
+ 20 more
Share this project:

Updates