Presentations¶
A running index of every ClawBio HTML deck, ordered newest first. Click a row to open the slides in a new tab; use arrow keys or space to advance.
| Date | Title | Context |
|---|---|---|
| 2026-09-14 | Validated skills are necessary but not sufficient for trustworthy agentic genomics | Manuel Corpas's Crick AI Club talk, recorded on Zoom on 14 September 2026. Science after pipelines: moving from manually connecting computational tools to directing scientific agents, with explicit methods, provenance and validation. Two pharmacogenomics figures show what validated rules improve and why genetic-input interpretation still needs separate validation. The 22-slide deck closes with composable skills, executable methods and community resources. Includes PDF and PowerPoint downloads, the full recording and podcast audio. |
| 2026-09-07 | Agentic genomics: Your science, made executable | Manuel Corpas's lecture for the INFLAMomx Training School at the University of Salerno. Real RNA-seq and single-cell examples introduce context engineering, inspectable skill execution, stopping rules, evaluation and equity as requirements for clinical-grade agentic genomics. Includes the downloadable PDF slides. |
| 2026-08-18 | Agentic Genomics: validated skills are necessary and not sufficient | Manuel Corpas's opening talk for the ClawBio + Nebius Hackathon at Impact Hub Berlin (18 August 2026). Fifteen slides adapted from the LatinOMICS deck and rebuilt on the Cell Genomics package submitted on 16 August 2026 (manuscript v72): 13,200 evaluations, eight frontier models, 110 CPIC Level A cases, five configurations. Carries three results the earlier decks did not have. Authoring the decision logic moves recommendation accuracy from 62.4% free and 55.1% RAG to 97.0%, with execution supplying determinism, bounded failure and abstention rather than accuracy. Retrieval inverts safety, raising lethal-class errors from 54 to 137 of 336, with 74 of the 83 extra errors on HLA loci and HLA-B*15:02 going from 0 of 48 to 45 of 48. And curated accuracy does not transfer: 96.4% on the benchmark against 68.2% on family WGS (n=4), 60.6% Iberian (n=93), 53.4% Peruvian and 37.2% East African. Closes on the corrupted-contract result (88 of 90 responses echoed deliberately corrupted rules, none reverted), the day's three challenges, and the verified Token Factory setup. Live ClawBio traction: 96 skills, 1,105 stars, 46 contributors, 8,182 Galaxy tools bridged. |
| 2026-07-14 | The Future of Transcriptomics in an Agentic World | Manuel Corpas's opening for the Londonomics RNA Ideathon at the Francis Crick Institute (14 July 2026). Fourteen slides arguing that the future of transcriptomics is not a smarter model but a trusted operating layer: skills that carry a versioned context contract, fail safely, and keep a chain of custody. Reproduces the published Klim and Melamed TDP-43 knockdown result (STMN2 collapse, GSE121569) through the ClawBio rnaseq-de skill, then a nine-model, five-provider invariance grid where skill execution collapses model-to-model variation, with the same collapse shown in DNA from the pharmacogenomics benchmark. Grounded in two Cell Genomics papers (the Agentic Genomics workflow perspective, live 21 July, and the trustworthy skill-library benchmark in revision) and live ClawBio traction (94 skills, 3,700+ tests, 43 contributors, 1,000+ stars). Closes on the thesis that the community should build the shared component every trusted RNA agent uses, plus a build-with-us slide (book + community). An appendix slide gives the five verbatim benchmark requests. |
| 2026-06-24 | Agentic Genomics in Practice (ClawBio × Genomic Intelligence) | Manuel Corpas's opening for the ClawBio × Genomic Intelligence webinar (24 June 2026, Zoom, 230+ registered). Fifteen-slide framing of the state of the art in agentic genomics before the Genomic Intelligence live model demos: the two waves of LLMs in biology, a four-condition definition of agentic genomics, the thesis that the bottleneck moves from producing analyses to validating them, and the validation bottleneck of silent plausible failure, including the real, publicly caught-and-fixed empty-file bug (a clean "all normal" report dosing 51 drugs). Grounded in the pharmacogenomics benchmark (44,550 scored evaluations; 80.6% free prompt → 95.5% skill reasoning → 100% skill execution), live ClawBio traction (~1k stars / 90 skills / 48 contributors), the counterintuitive result that the right guideline as free context made the model more dangerous, the ancestry accuracy gradient (72% European / 51% Latin American / 40% East African), and a three-tier validation framework (research / benchmarked / clinical). Closes on the conversational.clawbio.ai live demo and a Build-with-us slide (book Amazon QR + WhatsApp community). |
| 2026-06-18 | Build Your First Skill | Hands-on workshop deck for the KCL Hackathon. Covers why skills matter, working with coding agents, planning as a team, and getting from zero to an open PR in one afternoon. |
| 2026-06-18 | ClawBio Hackathon: Agentic Genomics @ King's | In-person hackathon at the SGDP Centre, King's College London (Memory Lane, SE5 8AF), 12:00 to 17:30. Covers the four conditions, the validation bottleneck, the three-tier framework, and the three-arm factorial benchmark, closing on five principles for responsible agentic genomics. Includes a live demo embedding conversational.clawbio.ai, where every answer is executed live by the ClawBio pharmgx-reporter skill on the openly published Corpasome. |
| 2026-05-27 | El Futuro de la Biología es Agéntico — Mecanismos arquitectónicos | Deck complementario al de LatinOMICS. Versión castellana del benchmark factorial de tres brazos (26.730 evaluaciones, 9 LLM de frontera, 110 casos CPIC Nivel A, tres condiciones: sin especificación / con la guía CPIC cargada / con especificación). 14 diapositivas que desarrollan tres mecanismos arquitectónicos por los que el contrato cumple los requisitos de calidad clínica que la simple recuperación de la guía no alcanza: resolución de la ambigüedad (gen, fármaco) (470/973 = 48,3% de las regresiones son respuestas sobre otro fármaco), imposición de la traducción (fenotipo → acción) (HLA-B*57:01 con fenotipo 1,000 pero acción 0,111), y determinismo frente a redistribución (Q3 se triplica de 122 a 341 cuando se carga la guía). |
| 2026-05-20 | El Futuro de la Biología es Agéntico (versión castellano) | Versión neutra en castellano del seminario UK DRI Imperial. Reutilizable para LatinOMICS y foros académicos en español. 14 diapositivas con las cuatro condiciones, el cuello de botella de validación, el framework de niveles, el benchmark factorial de tres brazos (26.730 evaluaciones) y la trayectoria de errores letales DPYD (77 → 26 → 0). Números actualizados frente al dataset bloqueado de tres ejecuciones. |
| 2026-05-20 | Hackathon ClawBio Madrid: Agentes de IA para Salud y Genómica | Hackathon presencial en Madrid (C. de Asura, 90), 12:00 a 18:00. Versión en castellano del taller de genómica agéntica, adaptada a los tres tracks del evento: nuevos skills, flujos de trabajo con agentes y auditorías de equidad con HEIM. Incluye programa, instalación de ClawBio, comunidad y pase al III Congreso Español de Medicina Genómica (21 y 22 mayo, misma sede). |
| 2026-05-06 | Reproducing Rare-Disease Blood RNA-seq Diagnostics at Sidra + QBB | Proof-of-principle demo for the Sidra Medicine Mendelian Team. Reproduces the diagnostic principle of the Genomics England NGRL paper (medRxiv 2026.03.19.26348811, 5,412 individuals, OUTRIDER + FRASER2 via DROP) as a new ClawBio skill (rdoutlier). Live terminal demo, embedded heatmap result on synthetic data, demo-to-production swap table for QBB controls + Sidra paediatric cases, and a 12-month pilot proposal. |
| 2026-04-29 | The Future of Biology Is Agentic | UK DRI seminar, Imperial College London. 25-min talk grounded in two source papers: the bioRxiv Perspective on agentic genomics (Corpas, Fatumo, Guio) and the Briefings in Bioinformatics benchmark of 9 frontier LLMs on pharmacogenomics (Corpas, Iacoangeli, Fatumo, Guio). Covers the four conditions, validation bottleneck, three-tier framework, and empirical evidence that a plain-text SKILL.md takes accuracy from 92.4% to 100% and eliminates DPYD lethal-case errors across all populations. |
| 2026-04-23 | AI Agents for Health: ClawBio Hackathon | In-person hackathon at University of Westminster. 85 attendees. Three tracks + TuringDB graph bonus challenge. |
| 2026-04-23 | Agentic Genomics: Lessons Learned | Talk given at ClawBio Hackathon #2: AI Agents for Health, University of Westminster. Personal account of one month contributing to ClawBio — what changed, what to watch out for, and what it means for agentic genomics. |
| 2026-04-22 | Agentic Genomics: Hands-on with AI for Variant Interpretation and GWAS | Virtual workshop, 743 registrants. Two sessions: Variant Interpretation + GWAS. Collaboration with Prof Segun Fatumo (QMUL/PHURI). |
| 2026-04-18 | The Open Standard for Biological AI Skills | 6th Shanghai Global Talents Innovation & Entrepreneurship Competition (Haiju Yingcai). Life and Health track. Community-driven skills, tiered validation, verification badge business model. |
| 2026-04-15 | Agentic Genomics: From Pipeline Automation to Autonomous Validation | SCGG Away Day, King's College London. Family WGS case study with tiered validation framework. YouTube · Podcast |
| 2026-04-02 | Agentic GWAS Workshop | GWAS + PRS walk-through for the ClawBio workshop series |
| 2026-03-28 | Exploring a 30x Whole-Genome Sequence | End-to-end WGS interpretation workshop |
| 2026-03-27 | Agentic Genomics Workshop | Intro deck for the Imperial / London workshops |
| 2026-03-27 | Agentic Variant Interpretation | Variant calling and interpretation hands-on |
| 2026-03-21 | Genomebook: Genotype-Driven Agent Reproduction | Research deck on reproducing historical figures from their genotypes |
| 2026-03-18 | ClawBio Hackathon: Agentic Genomics (19 Mar 2026) | Kickoff slides for the Imperial hackathon |
| 2026-03-12 | BioSkill Marketplace | The OpenClaw biology skills marketplace pitch |
| 2026-03-06 | ClawBio VC Pitch 2026 | Investor deck |
| 2026-02-23 | 10 Tips for Becoming a Top 1% AI User | Personal productivity talk on agentic AI workflows |
About these decks¶
Every deck is a self-contained HTML file, free to share and re-host. Use the arrow keys or space to advance, and F for fullscreen where supported. Speaker notes, when present, are toggled with S.
Want a workshop or talk for your team, meetup, or conference? Reach out via GitHub or clawbio.ai.