Geoff: The GEOspatial Fact Finder

Ask a question, get a map. A natural language interface for querying geospatial data.

🌐 Try Geoff Here! 🌎

---

📘 Latest

Geoff 0.2 is alive!

• Multiple geometry layers at once!
• New JSON Plan Architecture: Huge under the hood change. Instead of LLM generating SQL directly (inconsistently), the LLM now generates structure JSON plans and a new query builder interprets them into spatial SQL. Much more deterministic, much more reliable - although functionality is now added one structure at a time.
  • See Supported Question Types below for more
• Vector embedded examples & schemas: the descriptions of the datasets and the examples the LLM has to pull from are now in a vector database, so the question can easily be compared to them and only the most relevant info pulled in.

---

Motivation

How many questions can be answered with a map? Questions from urban mobility to planning to climate response and more can be answered with geospatial data, but can require complex spatial SQL queries and data cleaning that is non-trivial for non-technical users.

Geoff takes a prompt in natural language, converts it into a spatial SQL query, and displays the result on a map - shortening the time from question to insight for planners, geographers, and more.

Problem Statement: Planners, NGOs, activists, public employees and more have questions with geospatial answers but are often limited by the collection and querying of spatial data.

---

Features

• Natural Language Queries: Ask questions in plain English and get answers mapped to real geospatial data.
• Dynamic Schema Selection: Question is compared to vector descriptions of relevant datasets, ensuring efficient and accurate queries.
• Few-Shot Prompt Generation: Builds system prompts with cached examples to improve generation and reduce errors.
• Interactive Frontend: React + Tailwind interface with:
  • Prompt bar for natural language input
  • Table view of query results
  • Interactive map layer displaying queried geometries
  • Dynamic data dictionary to see available search criteria
• ETL Pipeline: Automated ingestion, cleaning, and transformation of multiple datasets, with an easy path to add more.
• Extensible Dataset Support: Current version supports 9 datasets; architecture allows seamless addition of new sources.

Dynamic data dictionary
Explorable results

Limits

• Geoff can only answer questions about things it has in its datasets!
• With new JSON plan architecture, Geoff can only answer questions that the query builder has been built for.

Architecture

Supported Question Types select

• select/1.0 Single layer
  • E.g. "Show fire stations built after 1980"
• select/2.0 Multi-layer
  • Returning multiple, unrelated geometries
  • E.g. "Show me parks and schools
• select/2.1.1 Spatial Filter
  • Using one geometry to filter another (showing both) (distance, intersects)
  • E.g. "Show bike lanes within 100 m of schools"
• select/2.1.2 Spatial Join
  • Joining two layers based on geometry (distance, intersects)
  • E.g. "Show each school and the address of its nearest fire station"
• select/2.2.1 Attribute Filter
  • Filter one layer based on the attribute of another, joined by an attribute
  • E.g "Show the ward that Allan Gardens is in"
• select/2.2.2 Attribute Join
  • Join two layers by attribute
  • E.g. "Join attractions with wards by matching ward_id"

aggregate

• Coming imminently

create

• Coming soon

route

• Planned

Application Architecture

flowchart LR
    subgraph Data["Database: Postgres/PostGIS"]
        ETL["ETL Pipeline<br>(ingest + transform)"] --> DB[("PostGIS Database")]
        Vector_DB[("Vector Database")]
    end

    subgraph Backend["Backend: Python + FastAPI"]
        Schema["Schema & Example selection (vector embed)"]
        Prompt["System Prompt Builder"]
        Query_Builder["Query Builder<br>(JSON → SQL)"]
        Exec["Execute SQL, Parse GeoJSON Layers"]
    end

    subgraph Ollama["LLM"]
        LLM["Generate Plan<br>(Natural Language → JSON)"]
    end

    subgraph Frontend["Frontend: React + Tailwind"]
        UserPrompt(["User Question"])
        Leaflet(["Layers on Leaflet Web Map"])
    end

    Frontend ~~~ Backend
    Backend ~~~ DB

    UserPrompt --Embed--> Schema
    Vector_DB --> Schema
    Schema --> Prompt
    Prompt --> LLM
    LLM --> Query_Builder
    Query_Builder --> Exec
    DB --> Exec

    Exec --> Leaflet

Loading

Tech Stack

• Database
  • Docker, PostGIS, PostgreSQL, pgvector, Python, SQL
• Backend
  • FastAPI, Ollama (local LLMs), OpenAI, Python, geoalchemy, sqlalchemy
• Frontend
  • React, Tailwind CSS, Vite, Leaflet.js, OpenStreetMap, Node.js
• Deployment & Infrastructure
  • Docker, Nginx (server & reverse proxy), Cloud VPS, VPN Tunnel

Datasets

The More Info section of the application has a dynamically generated data dictionary that will likely be more up to date than this section.

More datasets will be added in the future. Since the backend dynamically selects tables to send to the LLM, the only cost of increasing the number of datasets is more storage space.

• Ambulance/EMS Stations
• Attractions / Points of Interest
• Bike lanes
• Fire Stations
• Parking Lots
• Parks
• Police Stations
• Schools
• Wards

Roadmap

• ✅ Current:

  • Support Question Types
    • aggregate - Aggregated results and derived spatial features.
  • More Datasets
  • Automated JSON validation

• 🔜 Next:

  • Support Question Types
    • create - "LLM Generated" geometry
    • route - custom pathfinding
  • Place Hierarchy table
    • Better disambiguate between overlapping locations (E.g. "UofT" vs "University (Neighbourhood)" vs "University-Rosedale (Ward)" vs "Downtown" (Region), etc.)
  • User Feedback Pool
    • Add thumb up/down to each response. Add automated review of negatives. Use positives and corrections to negatives to continue building embedded vector example pool

• 🎯 Future:

  • Transparency & Reliability
    • LLM justifications
    • Plan/SQL visibility
    • User added context / clarification
  • Workspaces
    • Store multiple old layers, toggle visibility
    • Reference previous layers in new questions
  • QOL
    • Export Results (csv, geojson)
    • Sort/filter results
    • Open contributions, local setup instructions

Contributing

Suggestions & feedback are currently welcome. Open contribution is not currently available. If Geoff could help you or your organization, please reach out.

Getting Started

Instructions on setting up Geoff locally for custom use or additional data access will be available in the future.

Attribution

Datasets currently used by Geoff are sourced from:

• Toronto Open Data