LLM panel screens 15,034 records into 489 studies with 1.00 sensitivity
maier_ak · x · 2026-10-30
A demo of an LLM-automated systematic review pipeline: the first cycle screened 15,034 records, kept 8,699 after dedup, assessed 1,276 full texts, and included 489 studies, extracting platform, propulsion, and collective phenomenon fields. Snapshots are produced monthly under an MIT-licensed, forkable codebase.
Screening uses three open-weight LLMs voting per record, with inclusion requiring two votes. Calibrated on a 200-record human-labeled sample targeting sensitivity ≥0.95, the panel achieved 1.00 sensitivity and 0.98 specificity.
Related event: Living review uses LLMs to auto-update systematic reviews monthly(4 posts)→
More from coding & agent
- How Agno's 1,428-line Agent class stays thin: a façade pattern teardown — Mahmoud_Zalt · 2026-10-30
- Agent writes a script to remove slop comments — the script itself has 10+ lines of them — shadcn · 2026-10-30
- DataTalksClub and Nebius launch free course on production AI agents for generative media — Al_Grigor · 2026-10-30
- Faros report: AI speeds coding but QA time up 300% across 22,000 developers — FrankFelixAI · 2026-10-30
- CognitiveCanvas Turns AI Chat Memory into Auditable Graphs, Cutting Tokens 92.2% — JeremyCMorgan · 2026-10-30
- FontBox LoRA brings bbox-based intuitive text editing to FLUX.3 Klein — SillyLilithh · 2026-10-30