Skip to content

TaluGPT

TaluGPT is a natural-language map of Estonian local food.

Core trick

Retrieves farm and product records from Qdrant for a natural-language query and highlights the matching results on a Leaflet map.

Tags

Description

TaluGPT is a natural-language map of Estonian local food. 2,130 farms, producers, markets, shops and food events from the PTA organic registry, e-Äriregister, avatudtalud, kohaliktoit, EPKK and laadakalender, unified into a single searchable Leaflet map. Filter by county, food category (milk, eggs, honey, etc) or organic-only, then ask the embedded AI assistant Claude Opus 4.7 + Qdrant + n8n, questions like "Where to buy organic honey in Tartu?" and get an answer with the relevant farms highlighted on the map. Estonian and English in one space, but data based of Estonian farms, events etc. Live at https://talugpt.vercel.app
AARenor/talugpt★ 01 forksPythonCreated Apr 25, 2026 · last push Oct 8, 2026

README excerpt

# TaluGPT 🇪🇪

**Natural-language local-food discovery for Estonia.**

Ask in Estonian or English — *"Kus saab Viimsis toorpiima?"* or *"Where can I buy organic honey near Tartu?"* — and get a Claude-powered answer alongside a Leaflet map highlighting the relevant farms, markets, producers, shops, and food events.

> Live at **[talugpt.vercel.app](https://talugpt.vercel.app)**

![TaluGPT desktop](docs/screenshot.png)

---

## What it is

A web map of **2,272 deduplicated, geocoded** Estonian farms, producers, markets, shops, and food events, stitched together from the PTA organic farming registry, the Estonian Business Registry (e-Äriregister), avatudtalud.ee, kohaliktoit.maaturism.ee, EPKK, and laadakalender.ee.

Each pin carries the venue's kind, county, municipality, primary food category, products, certifications (e.g. *mahe*), contact info, tags, and source provenance. Filter chips on the side narrow the map by kind, county, food category, and organic certification. A floating chat widget answers free-form questions in Estonian or English by retrieving the most relevant venues from a vector database and letting Claude write the reply.

---

## What it's built on

| Layer | Tech |
|---|---|
| LLM | **Claude Opus 4.7** (`claude-opus-4-7`) — 1M context, prompt caching, streaming |
| Vector DB | **Qdrant** — collection `codex_drive_rag` |
| Embeddings | **Google `gemini-embedding-001`** — multilingual, called from n8n |
| Orchestration | **n8n** (self-hosted) — chat trigger →

Team