Mohammad TahaStatus: Live demoCo-built with partners
Mochi Kitchen
AI recipe finder
Describe a craving or list what's in the fridge, and Mochi finds recipes, merges shopping lists, ranks recipes by what you already own and plans your week. Built with Node, Express and Sequelize locally, and a Hono API on Cloudflare D1 in production — both sharing the same pure logic.
- Built with
- Node.jsExpressSequelizeJavaScriptHonoCloudflare WorkersCloudflare D1TheMealDB APIGemini API
- Type
- Personal project
- Source
- Public on GitHub
What was built
- Free-text search that tokenises and stems, then detects ingredients, cuisine, diet and calorie limits such as “under 500 calories”. repo docs
- Shopping lists that merge and de-duplicate ingredients and add units: 500 g + 0.7 kg = 1.2 kg. repo docs
- Accounts with short-lived JWTs and rotating, hashed refresh tokens with replay detection; Joi validation, rate limits and a CSP. repo docs
My part
- My Full Stack Development II project: 19 of 23 commits are mine, including the Worker placement fix. A collaborator added a Java microservice and a later theme. commits
Limits and open work
- Measured cold sign-up used 42 ms CPU — above Workers Free's documented 10 ms limit. It wasn't rejected in practice, but Cloudflare doesn't promise that. repo docs
Decisions worth knowing
Move the Worker to the data
D1 lives in Tokyo and a search makes dozens of small queries, each costing ~250 ms from another continent. Pinning the Worker next to the database took searches from up to 40 s to about 1–4 s.
repo docs
One logic, two runtimes
NLP, diet rules, grocery merging, scoring and the planner are pure modules shared by the Express app and the Worker, so behaviour is identical.
repo docs
Estimates say they're estimates
Calories are always shown as “approx.” or “AI estimate”.
repo docs
A closer look
500 g+ 0.7 kg= 1.2 kg