Research Workspace
A status view of the current research stack: what is running, what has been tested, and what still needs stronger validation.
storage Current Stack
| Component | State | Details |
|---|---|---|
| Knowledge Base | Running | 1,571 chunks from 46 papers. This is the retrieval layer behind the chat experience. |
| Vector Search (Upstash) | Running | 1,571 vectors, BGE-base-en-v1.5, 768 dim, cosine. Powers semantic retrieval over the corpus. |
| Chat Interface | Available | Source-grounded HD Q&A over the research corpus, experiment outputs, and trial data. |
| Daily Data Refresh | Scheduled | Pulls PubMed, ClinicalTrials.gov, and HDBuzz into the site data layer each day. |
| Publishing Guardrails | Enabled | Checks for required medical disclaimers before changes are shipped. |
| Analytics | Ancillary | Traffic measurement only. Not part of the research logic. |
science Experiments Run
| ID | Name | Date | State | What Was Tested | Report |
|---|---|---|---|---|---|
| EXP-001 | Abstract-Level HD Paper Pass | 2026-03-27 | Completed | Whether an LLM could extract structured targets, compounds, and candidate hypotheses from 22 HD paper abstracts. | View |
| EXP-002 | Full-Text Deep Analysis (16 papers) | 2026-03-30 | Completed | What changes when the model reads full papers instead of abstracts: 16 papers, 1.9M characters, deeper mechanism extraction. | View |
| EXP-003 | Gemma 4 vs Llama 3.1 | Designed | Planned model-comparison pass using the same paper set to compare extraction quality and hypothesis differences. | View | |
| EXP-004 | Somatic CAG Expansion Screen | 2026-04-07 | Completed | Focused screen on somatic expansion modifiers. 92 relevant papers were identified and synthesized into mechanistic intervention ideas. | View |
This table tracks research passes, not product launches. “Completed” means the analysis was run and published, not that the conclusions are validated.
smart_toy Automation Status
| Agent | Last Run | Total Runs | Latest Outcome | Schedule |
|---|---|---|---|---|
| Paper Scout | 2026-03-28 | 1 | Initial corpus build completed; future value depends on repeated runs and stronger validation. | Every 6h |
| Hypothesis Refiner | 2026-03-29 | 1 | Initial tracker pass completed; score histories exist, but evidence trails are still thin. | Daily |
| Social Watcher | Never | 0 | Scaffold exists, but this agent is not yet part of a trustworthy production workflow. | Daily |
| Digest Writer | Never | 0 | Planned summarization layer; not yet used as a stable public artifact. | Weekly |
lightbulb Hypothesis Tracker
| Drug | Target | Score | Evidence Record | Status |
|---|---|---|---|---|
| Tocilizumab | IL-6 / neuroinflammation | 80/100 | 1 score entry; no supporting or contradicting evidence logged yet | Exploring |
| Metformin | mTOR / AMPK / autophagy | 72/100 | 1 score entry; rationale captured in tracker | Exploring |
| Rapamycin | mTOR | 68/100 | 1 score entry; still an unvalidated mechanistic lead | Exploring |
| Riluzole | Glutamate excitotoxicity | 60/100 | 1 known negative/limiting note in tracker | Known/Tested |
| Lithium | TDP-43 / GSK-3B | 55/100 | 1 score entry; mechanism needs stronger support | Exploring |
Tracker rows are pulled from the current hypothesis file, not manually curated homepage claims. Scores are AI-generated prioritization signals, not clinical probability.
build Tooling
Claude Code
Development workflow
Ollama
Local model serving
Llama 3.1 8B
Experiment model
NVIDIA NIM
Chat inference
Jetson AGX Orin
Local compute
Upstash Vector
Retrieval layer
Sarvam AI
Language coverage
Vercel
Static hosting + APIs
PubMed API
Literature source
ClinicalTrials.gov
Trial source
HDBuzz RSS
News source
GitHub Actions
Scheduled refresh