Pilot track
AI in Software Development
Four evidence types and a comparison capstone on adopting AI tools.
4 paper labs + capstone
WHITEPAPER READING GYM
Read research at the source — before its lessons reach the books
Separate claims from evidence in 30 minutes and turn fresh papers into engineering decisions.
No summary before your own conclusion
30-minute lab
claim → evidence → decision
- 02Context
- 05First pass
- 10Evidence
- 08Critique
- 05Engineering memo
68sources in the catalog
5practices: 4 labs + capstone
localStoragelocal-only storage
Active reading
Your answer first. Expert calibration second.
Every stage asks you to commit to an answer before showing a reference. There is no score—only a self-check rubric and a memo you can use at work.
- 01
Open the original
Exact section, page, and figure locators take you to the primary source.
- 02
Form your version
State the research question, claim, and expected effect.
- 03
Check the evidence
Reveal the expert reference only after answering or explicitly skipping.
- 04
Build a memo
Decision, confidence, constraints, applicability, and next step—in 250 words.
Fresh reading
10-minute triage
Decide whether a paper deserves another 30 minutes: question, evidence, strongest claim, and your verdict.
BitsAI-CR: Automated Code Review via LLM in Practice
Run the triageResearch, not book chapters
Original-source catalog
One card per unique source. Videos, Telegram posts, and podcasts are companions—not duplicate papers.
Product experiments
AI-based IDE features at Google: how the effect on engineering flow is measured.
Randomized controlled trial
METR and experienced OSS developers: the effect ran against expectations.
Benchmark validity
SUSVIBES: functional code and secure code are two different outcomes.