Give AI the research. Keep the evidence.
A search tool for job listings: brief an AI researcher, compare what it finds and check each result against its original posting. It only finds and organizes listings. The public demo uses a broad fictional collection across different roles and industries.
Turn a broad search into something inspectable
A search result or promising title rarely tells the whole story. What does the job actually involve? Is the original opening still active? Are two results the same listing? What did the researcher find, and what is still unknown?
I built this tool to bring those questions together. AI can do the searching and reading; the interface keeps the resulting evidence available for a person to inspect.
The workflow: brief, research, review
- Define the question. Choose a broad example or write a research brief with the facts you want checked.
- Give it to an AI agent with web access. The downloadable brief asks the agent to find current openings, read original employer or ATS pages, and return source URLs, check dates, short supporting excerpts and open questions.
- Import the results. Bring the agent’s JSON file back into the tool. Every imported result starts as needing verification.
- Inspect and compare. Read the source evidence, follow the original link, search descriptions, filter the collection and collapse repeat listings.
The browser page does not run a model or search the web itself. Finding current information requires an external AI researcher; this demo provides the handoff and review surface.
Try the public example
The starting collection contains 16 fictional records across roles such as data analysis, design, operations, writing and research. Employers, excerpts and sample review states are invented. They illustrate missing evidence and repeated records without recreating a personal search.
Open Create an AI research brief, download the task, then inspect a listing’s evidence and unanswered questions. You can also import your own agent’s results. Imports stay in the current tab and disappear on refresh; the app does not upload them to a server.
What I built, and where AI fits
The original project combined AI-assisted research intake with a local Python pipeline for joining records, surfacing uncertainty and preparing a browser view. The portfolio version gives that workflow a generic brief-and-import interface, with independent filters and visible source evidence.
AI’s useful role is discovering and reading information that is absent from the initial collection. The interface handles the structured handoff and comparison. A model-produced quote, URL or summary still needs checking against its source.