Bid board that watches every procurement portal for you

Build me a bid board app that replaces the daily ritual of checking a dozen procurement portals by hand. The app opens on a pipeline of open solicitations laid out in four columns: New, Reviewing, Pursuing and Passed. Cards move between columns by dragging, and the column a card sits in is saved so the board looks the same when anyone on the team opens it.

Each card shows the solicitation title, the buyer or issuing agency, the estimated contract value, a fit score, and a deadline countdown showing days remaining until the closing date. The countdown turns red once a solicitation closes within seven days. Inside every column, sort cards so the closest deadlines sit on top. Clicking a card opens the full extracted detail: solicitation number, buyer, closing date, estimated value, the fit score with the agent's written rationale, and a link back to the original notice on the portal.

A Sources page is where the team registers what to watch. Each source is a portal URL plus the search terms that matter for that portal, and sources can be added, edited and removed. A separate capability profile page holds an editable description of what we can actually bid on: our services, contract sizes we can handle, certifications, geographies. Both the sources list and the capability profile persist in the app.

A "Scan sources now" button on the board fires a background agent that works each registered source in turn using Hyperbrowser. For every source the agent drives the portal like a person would: it runs the site search with the registered search terms and pages through the listing pages using a Browser Use task, waiting for the task result before moving on, since most procurement portals need form driven search and pagination rather than a plain page fetch. It scrapes the notice pages it collects, then AI extracts each notice into structured fields: solicitation number, buyer, closing date, estimated value, and the link to the notice. Where a portal is thin or a term needs broadening, the agent can also run a web search through Hyperbrowser to find notices it would otherwise miss.

The agent then scores each extracted notice against the capability profile, producing a numeric fit score and a short written rationale explaining why this is or is not a fit for us. Results land in the New column. Dedupe is a first class behaviour: before writing anything, match each notice by solicitation number against every card already on the board, in any column, and skip the ones already there. A rescan must never flood the board with duplicates or resurrect something the team already passed on. Show scan progress while the agent runs and a summary of how many new notices were added when it finishes.

When someone drags a card into Pursuing, the app creates a Notion page for that solicitation holding the full extracted detail, the agent's fit rationale and the source link, so the bid write up starts there rather than from a blank page. Store the resulting Notion page URL on the card so the team can jump straight to it. At the same time, post a short win theme summary for that solicitation to our bids channel in Slack, covering what the opportunity is, the deadline, and why the agent thinks we fit.

Dragging a card to Passed opens a small prompt asking for a one line reason for declining. Save that reason on the card and show it on the Passed column cards, so the team can review what we keep declining and adjust the capability profile or the source list accordingly.

Use Hyperbrowser for all portal browsing, scraping, AI extraction and web search, Notion for the bid write up pages, and Slack for the bids channel posts. The people using this are a capture or proposals team who open the board every morning, so the board itself has to be worth the open: scannable cards, obvious deadline pressure, and a trustworthy count of what is genuinely new.