ReMap · ری‌مپ

Persian-first Tehran metro router — a dark-mode MapLibre map, transfer-aware Dijkstra routing over the full Tehran network, saved places, and an LLM assistant that actually drives the map.

Live demo: https://reimap.vercel.app

🎬 Demo video: ReiMap.mp4

ReiMap.mp4

Features

• Routing that understands transfers. Dijkstra runs over (station, line) states, so changing lines costs a penalty and routes prefer staying on one line — exactly how riders think.
• Metro + BRT modes. The «خطوط و ناوگان» panel toggles each fleet on/off; the map layers, the routing graph, the search and the AI assistant all follow. Metro↔BRT transfers are real walking edges (≤800 m) with honest walk-time costs, so the router trades mode changes exactly like a rider would.
• Door-to-door endpoints. Origin/destination can be a station, a saved place, your GPS location, a map click, or a free-text address (Nominatim, bounded to greater Tehran). Non-station endpoints get first/last-mile walking legs with distance-based time estimates.
• AI assistant that controls the map. The LLM answers in Persian and emits <map>{...}</map> commands (route / show / save / nearby) that the app parses, validates and executes — streaming over SSE while command blocks are hidden from the chat bubble.
• Saved places. «خانه خاله», «دانشگاه»… stored locally, matched script-insensitively (alef/yeh/kaf variants, ZWNJ), and understood by the assistant by name.
• Cross-device sharing, no backend. Share your saved places as a link (#p=…, compressed into the URL hash) or a QR code — the data travels inside the link itself; nothing is stored on any server. Phone→desktop: send the link; desktop→phone: scan the QR. Incoming places merge with script-insensitive duplicate detection.
• Nearby POI search via Photon with brand queries («نزدیک‌ترین کافه لمیز»), distance-sorted numbered pins.
• Fully RTL Persian UI, Persian digits, under-construction stations dimmed on the map.

How it works

Routing core — src/lib/route-finder.ts

1. Graph repair. Upstream data misses some reverse edges; makeBidirectional mirrors every edge so lines run both ways.
2. State-space Dijkstra. A node is not a station but a (station, line) pair. Riding into a station on line A then continuing on line B costs the transfer penalty (MIN_PER_TRANSFER = 5 min) — the search naturally trades extra hops against transfers.
3. Legs reconstruction. The state chain is folded into contiguous per-line legs (the shared transfer station closes one leg and opens the next), which drive both the timeline UI and the map geometry.
4. Walking legs. Arbitrary endpoints snap to their nearest station (haversine); walking time uses 4.5 km/h.

Assistant protocol

user  →  "از دانشگاهم تا هتل اسپیناس"
model →  "الان مسیر دانشگاه تا هتل اسپیناس را نمایش میدهم."
         <map>{"action":"route","from":"دانشگاه","to":"هتل اسپیناس"}</map>
app   →  resolves both names (saved place → station → Nominatim),
         sets endpoints, draws the route, shows an action chip in the chat

The request body only carries { q, context, history }; the system prompt and every message role are composed server-side in src/app/api/ai/route.ts, behind a per-IP rate limit. Client code can never forge roles or stream arbitrary prompts to the model provider.

Getting started

Prerequisites: Node ≥ 20.

npm install
cp .env.example .env.local   # then fill in GROQ_API_KEY
npm run dev                  # http://localhost:3000

Get a free Groq API key at console.groq.com. Without it everything works except the AI assistant sheet.

Script	What it does
npm run dev	Dev server on port 3000
npm run build / start	Production build / serve
npm run lint	ESLint (flat config, next/core-web-vitals + TypeScript rules)
npm run typecheck	tsc --noEmit
npm test	Vitest — routing invariants incl. “every active station reachable from Tajrish”

Data & attribution

• Network data (stations, lines, graph, paths): collected and maintained by the author — regularly synced from public live-transit sources via a private data pipeline. Spot an outdated station? Open an issue.
• Map tiles: © OpenFreeMap © OpenMapTiles, data © OpenStreetMap contributors (ODbL).
• Geocoding: Nominatim (rate-limited client-side to respect its usage policy) · POI search: Photon by Komoot.
• LLM: Groq inference.

Contributing

Issues and PRs are welcome — especially data corrections for new stations/lines. Keep PRs small; CI (typecheck + lint + tests) must be green.

License

MIT. Network data remains subject to OpenStreetMap's ODbL where applicable.

---

ری‌مپ

مسیریاب مترو تهران — نقشه تیره با MapLibre، مسیریابی با الگوریتم دایکسترا و جریمه تعویض خط، مکان‌های ذخیره‌شده، و دستیار هوشمندی که خودِ نقشه را کنترل می‌کند.

قابلیت‌ها

• مسیریابی مبدأ تا مقصد درب به درب؛ ایستگاه، مکان ذخیره‌شده، موقعیت GPS یا هر آدرس متنی
• روشن/خاموش کردن ناوگان مترو و بی‌آرتی در پنل «خطوط و ناوگان»؛ نقشه، مسیریابی، جستجو و دستیار همه از حالت‌های فعال پیروی می‌کنند
• دستیار فارسی‌زبان که با دستورهای <map> نقشه را مسیریابی، نمایش مکان، ذخیره و جستجوی اطراف انجام می‌دهد
• زمان تقریبی سفر با احتساب پیاده‌روی اول و آخر مسیر و جریمه تعویض خط
• رابط کاملاً راست‌به‌چپ با اعداد فارسی
• اشتراک‌گذاری مکان‌های ذخیره‌شده بین دستگاه‌ها بدون سرور — لینک یا کد QR؛ همه داده‌ها داخل خود لینک است و ورودی‌ها با تشخیص تکرار ادغام می‌شوند

اجرا

npm install
cp .env.example .env.local   # کلید رایگان Groq را قرار دهید
npm run dev