Toolfork

Toolfork is a SaaS alternatives directory built less like a content site and more like a small data pipeline with a frontend attached. Under the hood, a Python enrichment pipeline pulls a base graph of tools and their alternatives from the SaaSHub API, then deep-scrapes each product's page (httpx + BeautifulSoup, retried with backoff) to extract structured pricing, platform support, feature lists, and screenshots the API alone doesn't expose. Every product's own outgoing alternatives are self-expanded on first enrichment, so the graph grows organically instead of staying limited to hand-picked seed terms. That scraped, verified data — never the model's own assumptions — is what gets handed to an LLM to synthesize a grounded verdict and FAQ set per product. Deterministic post-processing backstops the parts prompt instructions can't reliably guarantee on their own, because the goal is real content, not generic AI output dressed up as a listing. Nothing publishes on a timer or a human's say-so: each listing is scored against a 9-point completeness gate (real website, logo, categories, features, screenshots, platforms, priced tiers, a synthesized verdict, and 3+ FAQs) and only goes live automatically once it clears that bar. Anything short of it sits in a review queue instead of shipping half-finished. Every published page carries schema.org structured data (SoftwareApplication, FAQPage, BreadcrumbList, ItemList), and newly published pages are pushed to IndexNow the moment they go live rather than waiting on a crawl cycle. The whole pipeline runs unattended on a scheduled GitHub Actions cron — the catalog grows and re-enriches itself continuously, with no manual publishing step.