Stock Expert AI
An AI-based stock scoring and financial report analysis website that adopts a dual-channel rendering architecture, completely separating market data (numbers) from model commentary to ensure that page numbers always originate from data providers rather than model generation.
Timeline (last 90 days)
The founder gives two reasons: first, blind spots—the company's own agents are all built by the same person and share the same blind spots, so problems found by external agents are more valuable than internal checks; second, a public benchmark—anyone visiting the homepage can see the results.
The application process consists of five steps: validation, rate limiting, screening, decision and seating; all hiring decisions are made by humans, and the HR agent only provides recommendations.
The HR agent is another Claude agent running on a schedule; it reads each application, opens the public pages referenced in the application, checks the findings against the interview task, and gives a recommendation of hire, reject or uncertain; it may recommend hiring only when it can verify the findings itself.
The site treats everything in an application as untrusted data rather than instructions; the HR agent can only read public web pages and cannot post, send or modify anything, and its recommendations have no effect until the founder acts, in order to defend against prompt injection.
External agents can only work on public pages, are rate-limited to about one request per second, cannot log in, cannot submit other forms, cannot do load testing, and cannot touch private content.
Application validation includes strict per-role schema checks of the request (including length, URL format, links that must point to pages on this site, the 150-word limit for Explainer, etc.), and uses honeypot fields to block perfunctory bots.
The rate-limiting rules are: each agent can apply only once per role per day, each connection has a daily limit, and the entire job board also has a daily total cap.
Stock Expert AI is a US stock research website whose day-to-day operations are almost entirely carried out by 16 AI agents running on schedules, with only one human founder giving the final sign-off.
Stock Expert AI this week opened 10 'workstations' for AI agents, each workstation corresponding to a role, with the interview task being to complete a real task.
The ten open roles include: Bug Hunter, Accessibility Auditor, Fact Checker, Explainer, Translation Reviewer, UX Tester, SEO Scout, AI Answer Tracker, Link Scout, Question Hunter.
6 more timeline events
All Facts (16)
The founder gives two reasons: first, blind spots—the company's own agents are all built by the same person and share the same blind spots, so problems found by external agents are more valuable than internal checks; second, a public benchmark—anyone visiting the homepage can see the results.
50%UnverifiedThe ten open roles include: Bug Hunter, Accessibility Auditor, Fact Checker, Explainer, Translation Reviewer, UX Tester, SEO Scout, AI Answer Tracker, Link Scout, Question Hunter.
50%UnverifiedThe application process consists of five steps: validation, rate limiting, screening, decision and seating; all hiring decisions are made by humans, and the HR agent only provides recommendations.
50%UnverifiedThe HR agent is another Claude agent running on a schedule; it reads each application, opens the public pages referenced in the application, checks the findings against the interview task, and gives a recommendation of hire, reject or uncertain; it may recommend hiring only when it can verify the findings itself.
50%UnverifiedThe site treats everything in an application as untrusted data rather than instructions; the HR agent can only read public web pages and cannot post, send or modify anything, and its recommendations have no effect until the founder acts, in order to defend against prompt injection.
50%UnverifiedExternal agents can only work on public pages, are rate-limited to about one request per second, cannot log in, cannot submit other forms, cannot do load testing, and cannot touch private content.
50%UnverifiedApplication validation includes strict per-role schema checks of the request (including length, URL format, links that must point to pages on this site, the 150-word limit for Explainer, etc.), and uses honeypot fields to block perfunctory bots.
50%UnverifiedThe rate-limiting rules are: each agent can apply only once per role per day, each connection has a daily limit, and the entire job board also has a daily total cap.
50%UnverifiedStock Expert AI is a US stock research website whose day-to-day operations are almost entirely carried out by 16 AI agents running on schedules, with only one human founder giving the final sign-off.
50%UnverifiedStock Expert AI adopts a "dual-channel rendering" rule stipulating that the model is never allowed to generate any numbers, and all numbers are rendered as-is from the market data provider FMP.
50%UnverifiedIn Stock Expert AI's mechanism, news topic selection is not decided by the model; stories are picked from FMP data, including earnings calendars, stocks with the largest gains within two trading days (market cap floor of $1 billion), and sector themes, with the same stock not covered again within 7 days.
50%UnverifiedStock Expert AI's number gate extracts every number in the draft and compares it against the fact pack, allowing a rounding tolerance derived backwards from the written precision.
50%UnverifiedOnce an article is approved, the site stores the SHA-256 hash of the approved text and the hash of the evidence, and the article can be indexed only when both hashes still match.
50%UnverifiedThe principle of Stock Expert AI is that no matter what stage compresses or rewrites the evidence, verification must point to the uncompressed original file.
50%Unverified案例中公司内容接口为GET请求，返回JSON格式，无需参数，状态码code 200表示操作成功
50%UnverifiedStock Expert AI this week opened 10 'workstations' for AI agents, each workstation corresponding to a role, with the interview task being to complete a real task.
50%