Scout

Inspiration

Investors spend more time verifying startups than evaluating them. Research happens by hand across filings, press releases and scattered databases, while LinkedIn and X are full of self-promotion and claims nobody can check. Search surfaces marketing copy, not facts.

We wanted the ease of a social feed (follow companies, scroll updates, open profiles) with the trust of a research terminal. That became Scout: a daily Dispatch of startup signals (funding, hiring, product launches, partnerships), tuned to the sectors and stages you care about, where every real company links back to its source.

Scout: find the startups before they trend.

What it does

Sign in with any email. The login screen is its own full page and animates straight into the feed, with no reload. Returning users get a one-click Continue as ….
Six-step onboarding. Name and role, then sectors, startup types, stages, early signals and investors to follow. Choices are saved to your account, so they follow you to any device.
Live startup research. When onboarding finishes, Scout researches real startups that match your industries, signals and stages, using Akta's company and news data. They appear in your feed tagged REAL, with a link to the original article.
Honest labeling. Fictional demo companies are tagged AI. Real ones are tagged REAL. Figures are shown "as reported", and anything missing says Not reported instead of being filled in.

How we built it

Stack: Node.js + Express on the server, plain HTML/CSS/JS on the client, and Akta's MCP server for company data.

The front end. We started from a single-page Scout design with an editorial "Dispatch" look: cream paper, Georgia serif headlines, a rust accent. We kept the site the same and patched it with a build script, so every change is reproducible from the original design.
A connector. scout-connector.js sits between the page and the server. It owns the login screen and the transition into the feed.
Accounts and preferences. Email-only sign-in with an HMAC-signed, httpOnly cookie. Preferences are stored per user, and the server only accepts known preference fields. Writes must be JSON, which blocks cross-site form posts.
Research with Akta. We wrote a small JSON-RPC client for Akta's MCP server. Each Scout industry maps to Akta's industry taxonomy (AI → Artificial Intelligence & Machine Learning, Security → Cybersecurity, and so on). Each user signal maps to a news type (Funding → equity fund-raising, Product → product launches). Every candidate company is checked with Akta's free company search before it's added.
Sharing. We packaged a downloadable version (no keys, no user data) and a server-free copy that runs from a link. In the link copy, the connector swaps its server calls for in-browser storage and the cached research.

Challenges we ran into

Research on a small budget. Akta's company-list search is the ideal "find startups matching these criteria" tool. It costs $ c_{\text{list}} = 1.5 + 0.2n $ credits for $ n $ companies and needs at least 2.1 credits to run, but we had 0.81. We added a balance check before every paid call and a 12-hour cache, so the same search is never paid for twice.

News is noisy. Out of 21 articles across three industries, only 4 became real startups in the feed. The rest were stock-trade reports, CEO commentary, non-English pieces, investors, and even a government department that Akta listed as "private". One article about FirstWork's \$15M Series A listed Y Combinator as its main company. Our fix was to try every main company, headline names first, and reject investors, accelerators, banks and public bodies.
API details. Akta's MCP server expects the key in an x-api-key header rather than as a bearer token. Every company tool also accepts only a 7-character Akta ID, so names must be resolved first.
Getting the login flow right. We went through three versions: a separate page, then a pop-up over the feed, and finally a full login section that hands off to the feed with a transition. Opening the product always lands on login, and a live session becomes a Continue as button.
Accessibility. Re-rendering the onboarding window on every click dropped keyboard focus. We now restore focus to the control you used, move focus to each new step's heading, trap Tab inside the dialog, and expose selections with aria-pressed.

Accomplishments that we're proud of

A sign-in → onboarding → personalized feed flow that feels like one product, with no page reloads.
- Real startups matched to each user's criteria, every one traceable to a source article.
- Cost-aware research that degrades gracefully: when credits run out, users get a clear message instead of an error.

What we learned

Trust is a design problem. Labels like REAL, as reported and Not reported do as much work as the data itself.
Filters beat volume. A few strict checks (main topic, language, private status, not an investor) turned a noisy news stream into a short list worth reading.
Plan for limits. Credits, rate limits and plan tiers shaped the architecture as much as the features did.

What's next for Scout

Add Ask the startup, which answers only from verified data and says "Not in verified data" otherwise.
Email verification for sign-in, and alerts when a tracked company raises, hires or launches.