Journey
How the project evolved from a curiosity-driven build into a more useful research workspace.
Started with the question: can open tools make HD research easier to follow?
The first step was not model prompting. It was landscape work: understanding the HD pipeline, the public data sources, and what was missing for non-specialists and builders. That led to a stack for papers, trials, hypotheses, and site generation running locally on a Jetson.
Moved from collection to experiments
Once the data pipeline existed, the next step was to test what LLMs were actually good for: extracting structure from papers, surfacing targets, and proposing hypotheses worth human review. The important part was publishing both the outputs and their limitations.
Read the report arrow_forwardTurned the experiments into a usable interface
The project became more than a backend once the dashboard, chat interface, learning path, and automation loop started working together. That changed the purpose of the site: not just running experiments, but helping someone learn the field, inspect the evidence, and stay current.
What needs to get better next
- Make chat more visibly source-grounded and easier to verify
- Improve the learning path so newcomers can ramp faster
- Strengthen automated gathering and hypothesis provenance
- Add more validation around what gets published to the site
Build the loop. Make it legible. Improve it in public.