Ledger Sweep
Reconciles payment records against the ledger overnight and flags only genuine mismatches. Built for Quiet Hours — build an agent that does useful work while its user is asleep. Reconciliation across two sources with a confidence threshold. The interface is deliberately narrow: one path, done properly, rather than a broad surface that half-works at the deadline.
Gallery
Screens
Ledger Sweep — main view
Ledger Sweep — result state
On the clock
What shipped, what didn't
Completed in time
Reconciliation across two sources with a confidence threshold.
Unfinished / faked
Threshold is hand-tuned; no learning loop.
Obstacles
The first approach needed a background worker I didn't have time to deploy, so I moved it inline at minute 14.
Lessons learned
The demo path is the product. Everything not on it can wait for next time.
Process
Prompting approach
Started from a single system prompt describing the user and their constraint, then iterated with the assistant on the Ledger Sweep data model before writing any interface code. Full prompt log is in the repository under /prompts.
Fairness
Disclosures
Declared by the builder at submission time, visible to judges before scoring.
Judging
Public feedback
Judges' private notes stay private — this is what they chose to say publicly.
Elena Voss
VP of Engineering, Harbor Compute
The scope discipline here is the whole story. One path, finished, demoed without a stumble. That is what a 30-minute build should look like.
Hana Kobayashi
Principal Research Engineer, Independent
Genuinely useful. I would install this. The limitations section is honest, which makes me trust the rest of the submission more.
Claudia Reyes
Staff Engineer, Vantage Analytics
Excellent use of the model — it is doing actual work here, not decorating a form. The failure case you showed earned you points rather than costing them.
Discussion
0 comments
No comments yet.