Skip to content
← All AI workflows

Reuse a research engine for briefs and topic monitoring

Thrive Capital's Puck uses one internal research engine for interactive questions, background topic updates and company briefs. It combines firm records with approved external research. Puck and its task-coordination system, Polymer, are proprietary architecture patterns, not public packages.

From the source

Public video page from Linus Lee / Toronto Machine Learning Society (TMLS)
Linus Lee / Toronto Machine Learning Society (TMLS) · Captured 2026-10-14Read the original

How it works

  • The talk account reports over 80% background workload and approximately five-minute median and twenty-minute 90th-percentile task latency. The sample has no calendar dates or task denominator.
  • The separate operator post reports about 10 billion October tokens across thousands of tasks. It is not the same seven-day sample.
  • The video-page screenshot does not verify the talk’s figures. They remain operator-reported, with no independently reproduced outcome.
  • The architecture combines interactive questions and background research. No business improvement for Thrive or public Polymer implementation is established.

Workflow guide

  1. Choose fixed research questions, monitored topics and companies. Approve access to notes, email, calendar context, financial and investment records, and public or partner sources.

  2. Plan reusable retrieval and synthesis components in an isolated pilot. Treat Puck and Polymer as private architecture references, not downloadable software.

  3. Use the same components to draft answers, topic updates and stored company briefs. Cite sources and separate confidential firm material from external research.

  4. Keep traceable records for retrieval and draft generation. Check the fixed task set after source changes to detect silent regressions.

  5. Have staff verify briefs and material alerts before external communication or investment decisions. Require separate scheduling approval. Publish nothing automatically.

  6. Compare the pilot with manual research on the fixed tasks. Record acceptance, citation accuracy, review minutes, alert precision and cost per accepted brief separately from token use.

Sources and downloads

Discuss this workflow →

Book a call

A free 30 minute call. Tell us where you are with AI, what your IT allows, and what management needs to see. We'll tell you honestly what we would do first.