GC AI · Jev Hack Day · Sep 29, 2026

Jev Hack Day: what we built and what we learned

Every project and experiment from the hack day, mapped as reusable design patterns. Click any card for the query, results, and links to the PR, Loom, and write-up.

0projects and experiments
0.1 to 0.5 sper call vs 2 to 25 s
10x to 100xcheaper than the LLM
Where it did well vs. struggled ↓
Diagrams:codeJevLLMpersoninput / output
What the UI placements look like (mockups from the demos)
Can you review this NDA and flag anything risky?
＋ NDA Risk Review
Skill match: pill as you type, ~220 ms.
Going forward, keep answers under 200 words.
Save as a preference? · Length
Preference memory: Jev screens, Sonnet confirms.
Rent abates for 12 months1. Tenant may terminate at will2 ✕. Renewal is automatic3 ?
Citation checks, as the answer streams.
Act nowBreach notice · 2 days
This weekE-sign · 5 days
Needs a lookInvoice (injection)
Inbox triage: 9 questions per email.
Evidence

Where Jev did well vs. where it struggled

Task type Verdict Evidence Projects

Speed and cost vs. the comparison LLM

Times faster and cheaper, log scale.

Quality head-to-heads

● Jev  ● comparison  ● Jev + fallback. Axis 50% to 100%. Small samples.

Sources: hack day submissions, #jev-hackday and related Slack, the code in each app-gc-ai PR, and Chris Dzoba's jev-explore FINDINGS.md. Looms, Drive videos, and the private standalone repos were not readable, so those projects rely on their write-ups. Numbers come from small, often synthetic sets, usually one run: direction, not proof.