ChillClock

Inspiration

In a busy kitchen, a temperature log tells staff what a thermometer measured, but it does not necessarily help them see whether a batch is heading toward a missed cooling deadline. We built ChillClock to make those readings more useful: estimate the likely cooling timeline, explain when a batch may be at risk, and give staff practical next steps to consider.

What it does

ChillClock is a food-cooling decision-support prototype for kitchens, caterers, and food-service teams. Staff enter two temperature readings from the same batch and the elapsed time between them. The app uses a simplified Newton-cooling model to estimate when the batch may reach 70°F and 41°F, then compares those estimates with prototype reference targets.

The dashboard also takes context such as food depth, cover or ventilation, and cooling method into account when presenting rule-based suggestions. Staff can log a corrective action and review a local audit trail that links records with SHA-256 hashes. A Model Context Protocol (MCP) server exposes predict_cooling and get_cooling_guidance, and the dashboard has a "Run via MCP" path so the tool workflow can be demonstrated.

How we built it

We built ChillClock with TypeScript and Node.js, using Express for the server, the official Model Context Protocol TypeScript SDK for the MCP tools, and Vitest for tests. The cooling engine is separated from the dashboard and coaching rules. A local JSON Lines ledger stores prediction and corrective-action events, while a SHA-256 hash chain makes unexpected changes to the stored sequence easier to detect.

The dashboard includes example scenarios so a reviewer can try the workflow quickly instead of waiting through an actual cooling cycle.

Challenges we ran into

One challenge was making a useful estimate from only two readings. Real food cooling is affected by more than temperature and elapsed time, so a simple model must not be presented as a guarantee. We made the assumptions and limitations visible and labeled the reading-interval indicator as a heuristic rather than a statistically validated confidence score.

Another challenge was getting the MCP workflow and the regular dashboard path to work consistently, while ensuring that predictions and corrective-action logs could be checked in the audit trail. We verified the local typecheck, test suite, and MCP smoke test during development.

What we're proud of

• A working cooling-estimate flow that shows projected times to two target temperatures.
• A dashboard that turns a flagged estimate into practical, context-aware suggestions.
• An MCP server with separate prediction and guidance tools.
• A local SHA-256 hash-chain ledger and a verification view.
• A quick demo workflow that lets reviewers explore the full interaction in a short session.

What we learned

We learned how to connect a deterministic calculation engine to an MCP tool interface, and how important it is to expose assumptions when applying a simplified model to a real-world safety problem. We also learned that an audit hash chain can help reveal changes, but local storage and hashes alone do not make a record system tamper-proof.

What's next

Next steps could include integrating calibrated Bluetooth probes, tracking multiple batches and sensors, and conducting validation with real cooling data and qualified food-safety guidance before any operational use. A hosted deployment and a real voice-assistant integration would also need to be built separately.

Important limitations

ChillClock is a prototype and cannot determine whether food is safe. Its Newton-cooling estimates are approximate, depend on assumptions that may not hold, and must not replace physical temperature checks, approved corrective procedures, or applicable local requirements. The displayed 120-minute first-stage and 360-minute total targets are prototype reference values, not a determination of the rules that apply to a particular operation.

The coaching in the current version is rule-based; AWS Bedrock is not integrated at runtime. The MCP tools are available for demonstration, but ChillClock is not a published Alexa skill or a deployed voice assistant.

Built With

• css
• express.js
• food-cooling
• food-safety
• github
• html
• javascript
• mcp
• node.js
• sha-256
• typescript
• vitest
• zod