Inspiration
Thirty years of building PCs — every machine since Windows 95 hand-assembled, never store-bought — and the same scene on repeat: people paying for performance they will never feel, while overlooking the one cheap part (usually storage) that is actually holding them back. Every configurator on the internet stands on the overselling side. Nobody had built the tool that argues the other way.
What it does
You snip two screens Windows already has — Settings › About and the drive optimizer — paste them in, and say what you use the machine for. MOTTAINAI tells you which parts to keep, which parts are already far beyond what your use needs, and the one part, if any, genuinely worth replacing — sized to the need, not to what you own. For Windows 10 machines it lays out the paths that don't involve buying a new computer. The answer it is most pleased to reach is ¥0.
How I built it
Google ADK (TypeScript) runs two perception agents concurrently — one reads the hardware from the images, one interprets the workload — and they join exactly once into a deterministic judgment engine. No model touches a verdict or a price: core/verdict.js is plain tested code, 127 assertions. Gemini 3.7 Flash does the two jobs models are good at — reading images (even Japanese-language Windows screens) and putting a finished result into words — and the narrator is forbidden from introducing a number the engine did not produce. Deployed on Cloud Run.
Challenges I ran into
ADK's static graph fires a downstream node once per incoming edge, so joining two perception agents without running the merge twice took dynamicEntry. The harder class of bug was my own code lying with money: an early version printed "solved for free" while quietly needing ¥15,800 of RAM, and a later one asserted TPM state it had never read. Both are now guarded by tests that fail if a headline overclaims.
Accomplishments that I'm proud of
A tool that recommends not buying — no affiliate links, no product placement, no place for them. 5,948 CPUs and 3,013 GPUs looked up against Microsoft's actual Windows 11 support lists, with null answers that stay null and say why, instead of interpolated guesses.
What I learned
Deciding in advance what the model is not allowed to decide turned out to be the center of the design. Everything that touches money runs in ordinary code that a test can pin down.
What's next for MOTTAINAI
Laptop-specific judgment (screen, camera, battery — the parts this version deliberately left out), keeping the price data fresh, and running it as a permanent site.
Built With
- cloud-run
- express.js
- gemini
- google-adk
- node.js
- typescript
Log in or sign up for Devpost to join the conversation.