DatePilot A little more together

A developer built DatePilot, a full-stack private AI date planner for couples that combines a deterministic Python planner with an open-weight Gemma model served through Ollama. The app, built with React, Vite, Tailwind, FastAPI, Pydantic, and SQLite, keeps the LLM limited to interpreting taste preferences while code handles prices, schedules, and constraints, and it supports local or swappable hosted open-weight endpoints.

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend https://dev.to/challenges/hacktoberfest-weekend-2026-10-01 also i have done this with my friend sudharshan also u connect with gemma local models What I Built I built DatePilot, a private AI date optimizer for couples. The idea came from a simple real problem: planning a date for two people is weirdly hard. One person likes quiet cafes, the other wants something outdoors, both have a budget, timing, food limits, travel limits, and nobody wants to turn the date into a logistics spreadsheet. DatePilot lets two partners enter their preferences separately, then creates one shared date plan that fits both people. It finds overlapping tastes, keeps private answers hidden, and generates a real itinerary with: It is built for someone who wants the date to feel thoughtful without giving away the surprise or exposing private taste data. Live demo: Backend API: https://datepilot-api.onrender.com https://datepilot-api.onrender.com Video demo: GitHub repo: https://github.com/sudarsan2507-hue/DatePilot https://github.com/sudarsan2507-hue/DatePilot DatePilot is a full-stack app with a deterministic planner and an open-model AI layer. The frontend is built with React, Vite, and Tailwind , designed mobile-first so Partner B can comfortably open the invite link on a phone. The interface keeps the tone warm and romantic, but the workflow is practical: create a session, enter preferences, review taste cards, compare overlap, then generate a plan. The backend is built with Python, FastAPI, Pydantic, and SQLite . The planner itself is pure Python. It does not let the LLM decide prices, schedules, or constraints. Instead, deterministic code checks the real rules: The AI is used only where it is useful: understanding taste. The app is designed around an open-weight model wrapper using Gemma through Ollama , with environment variables for model name and base URL so it can run locally or be swapped for a hosted open-weight endpoint. The memory layer is SQLite-backed through a separate interface so future taste learning can be plugged in cleanly. I also added optional no-key public APIs: The venue planner currently works for India in the UI, with Tamil Nadu cities supported internally. Open innovation matters because this app is built around private taste. A date planner is not just asking for generic preferences. It can touch food habits, comfort zones, location patterns, aesthetics, budget comfort, and personal likes or dislikes. That data should not have to leave the user’s control just to get a thoughtful recommendation. Using an open-weight model makes DatePilot feel different from a normal closed API app: The important part is that the AI does not become a black box that secretly decides the date. The model helps understand messy human taste, while transparent code handles the math, constraints, and safety checks. That combination is exactly why open models matter here: they make AI personal without making it invasive. I used an AI coding agent codex during the build to iterate on the planner, frontend, deployment, API integration, and demo video. The agent helped debug real issues like failed plan generation, partner submission state, Render deployment behavior, and mobile sharing. I am entering: