The project is an attempt to automate the full market intelligence workflow, using fashion as first vertical, probably then to expand.
Process trigger is hourly industry news collection and selection, via Tavily and RSS feeds. Business selection rules dedupliaces and pick topics worth investigating.
Then an agent with industry main KPIs list is in charge to figure out which metrics could help explain what's happening, given a constraint of viable sources (mainly web scraping), and determines what data to collect.
Firecrawl handles scraping, other sources are internal. System structures data, run calculations, generate BI charts and insights, and publish the results.
The whole pipeline runs automatically, from identifying a market question to producing a data-backed analysis.
Additionally users may request custom reports, but this is non-essential to the service, which can be accessed without log in and without paid plans
The tricky part is - despite of the promise of most scraping tools like firecrawl and others, to affectively close the loop of data acquisition. not easy (we had to integrate it with data marketplaces with humans in the loop).
We decided not to have an llm chat UI, not 100% sure of this choice, but - as of today - it is not present.
Project is live at [https://tfb.databoutique.com](https://tfb.databoutique.com), with public analyses you can explore without signing up.
The project is an attempt to automate the full market intelligence workflow, using fashion as first vertical, probably then to expand.
Process trigger is hourly industry news collection and selection, via Tavily and RSS feeds. Business selection rules dedupliaces and pick topics worth investigating. Then an agent with industry main KPIs list is in charge to figure out which metrics could help explain what's happening, given a constraint of viable sources (mainly web scraping), and determines what data to collect. Firecrawl handles scraping, other sources are internal. System structures data, run calculations, generate BI charts and insights, and publish the results.
The whole pipeline runs automatically, from identifying a market question to producing a data-backed analysis.
Additionally users may request custom reports, but this is non-essential to the service, which can be accessed without log in and without paid plans
The tricky part is - despite of the promise of most scraping tools like firecrawl and others, to affectively close the loop of data acquisition. not easy (we had to integrate it with data marketplaces with humans in the loop). We decided not to have an llm chat UI, not 100% sure of this choice, but - as of today - it is not present. Project is live at [https://tfb.databoutique.com](https://tfb.databoutique.com), with public analyses you can explore without signing up.
Feedback apreciated.