How I built a free account research and outbound engine
A look under the hood of the account research tool: the free data sources, the no-fabrication design, and the infrastructure that runs the whole thing at zero cost.
I built an account research and outbound engine that takes a company domain and returns a sourced brief, a fit score, the people to contact, and a full outbound sequence. It works for any kind of business, runs on public data, and costs nothing to operate. Here is how the whole thing is put together.
The shape of the system
Every request flows through three stages: gather public signals, classify and synthesize, then produce the outreach. Nothing is invented along the way.
The data layer
The hard part of research is not writing, it is knowing real things about a company. I pull from sources that are free and that I can cite.
- The company's own site. The richest universal source is the structured data sites publish about themselves. Parsing schema.org and JSON-LD gives a real address, opening hours, price range, ratings, founders, and social profiles, the same for a restaurant as for a software firm.
- Wikipedia and Wikidata. Verified facts like founding year, headcount, and industry, confirmed against the company's official website so the wrong entity never gets attached.
- News, sitemaps, and hiring. Recent headlines for timing, the sitemap for catalog and location scale, and careers pages for growth signals.
For software companies it also reads GitHub and tech-stack signals, but those stay one input among many rather than the whole picture.
The no-fabrication rule
The single most important design decision was that the model never invents a number. Every figure in a brief traces back to one of the sources above. When the outreach wants to cite a result the seller has not provided, it writes a bracketed placeholder for the human to fill in rather than making one up. A brief you cannot trust is worse than no brief, so honesty is enforced in the prompt and checked in the output.
The infrastructure
The stack is deliberately boring and free.
A web app serves the interface and the API routes do the work on the server: fetch the public sources in parallel, classify the business, and call a free language model only to synthesize the qualitative parts, never the facts. It is hosted on a free tier, so the running cost is genuinely zero.
Security, because it fetches user input
The moment a tool fetches a URL a stranger gives it, it has an attack surface. So the engine refuses requests aimed at internal or private addresses, rate limits each visitor, caps input sizes, and ships a full set of security headers. A small free tool still deserves to be built like it matters.
What it adds up to
The result is a research-to-outreach engine that is honest, works for any business, and costs nothing to run. The next two posts cover why I bothered building it myself, and how the lead discovery finds real companies without paying for data.