What we’re built on.

App Router with server components. The marketing pages and the dashboard are rendered on the server, so a page of scoring evidence arrives as HTML rather than as a spinner followed by three fetches.

Only what genuinely needs interaction ships JavaScript — the opportunity filters, the pipeline runner, the preview cards. Everything else stays a server component and costs the browser nothing.

Scoring weights, database rows, and API responses share one set of types. Change how an opportunity is scored and the build fails, rather than a customer's ranked list quietly changing meaning.

One set of tokens defines the palette, radii, and shadows for both themes. Nothing hardcodes a colour, which is why dark mode is a variable swap instead of a second stylesheet.

Runs the rendered pages and caches them at the edge. Preview deployments mean a change to the scoring copy can be read in context before it reaches anyone.

Every table is scoped to a workspace by a policy inside the database, not by a WHERE clause someone has to remember. A query that forgets the workspace returns nothing instead of returning someone else's opportunities.

Candidates, enrichments, opportunities, publishers, and contacts are genuinely relational, and ranking is a query over them. Keeping the evidence in a real database is what makes a result inspectable months later.

Trigger.dev

Runs the long jobs.

Discovery, enrichment, qualification, and contact discovery outlive any HTTP request. They run as durable tasks that survive a deploy and retry a flaky provider instead of losing an hour of work.

Dice judges an article by its content rather than its title, which means something has to fetch the page and turn it into clean text first. This is the step most tools skip.

Content relevance, placement suitability, and the competitor gap come from a model that has read the page. The weights that turn those into a score are ours, deterministic, and published on the home page.

Hunter

Supplements contact discovery.

Public author bylines and contact pages come first. Hunter runs only after a page is qualified, so credits are never spent on a candidate that the hard checks were going to reject anyway.

Handles the paid trial and the monthly plan, and tells the app what someone is entitled to through a webhook. Access is derived from that, never from a flag set by hand.

Sign-in links, receipts, and the notification that a pipeline run has finished. A plain API and sane deliverability defaults, which is what keeps a small sender out of the spam folder.

Outreach goes out from your own address, not from a shared pool. OAuth means the connection can be revoked from your Google account at any time, without going through us.

The same arrangement for Microsoft 365 and Outlook accounts, through MSAL. IMAP and SMTP remain available for everything neither provider covers.

Server rendering, the API routes, and the background tasks all run on it — one language across the request path and the pipeline.

Strict by default: a package you did not declare is a package you cannot import, which catches a whole class of “works on my machine” before it ships.

Model output, webhook payloads, and request bodies are parsed rather than trusted. A malformed provider response fails at the edge with a readable error instead of becoming a null halfway through a pipeline run.

The scoring, ranking, and hard-check logic are pure functions with unit tests, so the numbers on this site can be derived from the same code that produces them in the product.

Rules for React and Next.js specifically — the ones that catch a client component that did not need to be one, or a state update that will cascade.

One consistent stroke weight across the marketing pages and the dashboard, tree-shaken so a page ships only the icons it draws.