Ecosystem Mapper: an AI agent that visualizes technology landscapes

I spend a lot of time trying to understand technology landscapes — which tools exist in a given space, how they relate to each other, what the main categories are, and where the gaps lie. This kind of mapping is essential for making informed decisions about what to use and what to build, but doing it manually is extraordinarily tedious. You search GitHub, read blog posts, scan Product Hunt, cross-reference Hacker News discussions, and eventually build a mental model that's incomplete and immediately starts going stale. I wanted to see whether an AI agent could automate the discovery and categorisation part, and the result is Ecosystem Mapper — a Python tool that takes a single keyword and produces a structured taxonomy of the technology landscape around it, complete with visual maps.

From keyword to landscape overview

The pipeline works in two stages. In the first, the agent collects data from multiple sources: the GitHub API for recent repositories matching the keyword, and Tavily web search for broader coverage of commercial tools, blog posts, and documentation. It then feeds everything through Google Gemini via OpenRouter to cross-analyse the results and produce a structured taxonomy — categories, subcategories, and representative examples with brief descriptions and links. The quality of this stage impressed me more than I expected; Gemini is genuinely good at identifying patterns across messy data sources and producing coherent categorisations. The second stage takes that taxonomy and generates visual ecosystem maps in the style of analyst market maps — the kind of visual overview where you can see at a glance how a space is structured and where the major players sit. The whole thing is designed to go from a single keyword to a comprehensive landscape overview with minimal human intervention.

What I've learned using it

The tool is built in Python with a modular architecture separating data collection, taxonomy analysis, and visualisation into distinct components. I've been using it to map spaces I'm actively exploring — MCP servers, voice technology tools, personal AI architectures — and it's become genuinely useful for getting quick overviews before diving deeper. The output quality depends heavily on keyword specificity: broad terms like "AI" produce unwieldy maps, while specific terms like "MCP gateway" or "Hebrew speech recognition" generate tight, actionable taxonomies. The data collection stage is working well; the visualisation stage is still in progress and experimental. The project is on GitHub and I'd welcome contributions, particularly on the visualisation pipeline.