R&D Amplifier
Self-hosted research intelligence

Science publishes 10,000 papers a week. You need twelve.

R&D Amplifier reads the literature end-to-end on one local machine — scores every paper, maps the field, and rewrites the signal into articles worth your evening.

0
papers analyzed
0
articles published
0%
avg eval
0%
local inference
scroll
arXiv bioRxiv medRxiv NeurIPS ICLR Nature gemma4-26b-a4b qwen36-35b-a3b vLLM · NVFP4 DGX Spark Geist Mono Instrument Serif
Most discoveries die in PDF — unread, unranked, unconnected. We built an *instrument* that reads everything, keeps only signal, and prints its provenance on every page: the model, the persona, the score. No black boxes. No feeds tuned for outrage. Only the research that deserves your *evening.*
The Forge — six stages, two gates
01 / 06
01 — INGEST

Ingest

Pull the source PDF and metadata from arXiv, bioRxiv and peers — on schedule, in your domains.

02 — PARSE

Parse

Full text, figures, references — reconstructed into clean structure the models can actually reason over.

03 — SCORE

Score

Novelty, applicability, quality. Anything below the signal gate never reaches your feed.

quality gate
04 — WRITE

Write

A persona-driven rewrite: engineering deep-dive, critical review or narrative — matched to the paper.

05 — EVALUATE

Evaluate

Every draft is judged against the source. Below the gate it is regenerated, not published.

quality gate
06 — PUBLISH

Publish

Site, knowledge graph, Telegram digest — with the model and score printed on the page.

Knowledge graph

Every paper leaves a trace.

Methods, models, findings — extracted from every analyzed paper and linked into one navigable map of your field.

0 entities 0 relations 0 linked articles

Research, amplified.

A morning digest of everything that cleared the gates — in Telegram, before your first coffee.