Seven departments at once
Each department is its own serverless call fired from the browser, so verdicts land as they finish. The Brain call waits for all seven to settle, which makes a committee two waits instead of eight.
Put a capital decision to seven AI departments, from Finance to Sustainability, and a Company Brain that rules on it. Then stress-test the numbers and drill the cases consultants actually use.
Finance argues from the hurdle rate. Legal holds a narrow veto. Operations wants a Class 3 estimate before anyone signs. Each department is its own Claude call with its own persona and its own blind spots, and all seven run at once. The Company Brain rules only after every department has reported.
The real app. Hover a department to fly in and click it to meet its agents. Pick a sample decision and press Convene: with no API key on the server it replays a sample committee.
Seven departments · one Company Brain
Hover a department to fly in · click to meet its agents · results below
An assumptions sheet with blue inputs, a live WACC slider with Gulf reference bands, and NPV, IRR, MIRR, payback and LCOE. The tornado swings each driver between real engineering lows and highs, so you can see which ones actually move the answer.
Market sizing, profitability and M&A on the strategy track; LCOE, debottlenecking and district cooling on the technical one. Ask for data one exhibit at a time, take a framework hint without the answer, then compare your issue tree with the model structure.
Serverless functions hold the API key and return schema-validated verdicts. The committee's lean is plain arithmetic in code, because models are unreliable at weighted sums. Every tariff, discount rate and case figure traces back to the research notes in the repo.
Damodaran KSA CRP 0.78%SAR 0.20/kWh tariffAACE Class 3RAPID vetooutput_config.format
Each department is its own serverless call fired from the browser, so verdicts land as they finish. The Brain call waits for all seven to settle, which makes a committee two waits instead of eight.
Blue Strategy and violet Marketing sat side by side at ΔE 3.3 under deuteranopia. Reordering the ring so similar hues never touch lifted the worst neighbouring pair to 7.4, and every hub carries its name as a backup.
Real clicks kept missing the tiny hubs, so each one now has an invisible pick sphere. A leftover constant had also thrown inside the render loop and silently swapped in the flat map; the fallback now logs why it fired.
The solar preset reproduces the research sanity case: NPV +$1.71M, IRR 11.2%. Writing the tests also caught my own wrong expectation: the textbook IRR is 8.9%, not 10.65%, because NPV at 10% is already negative.
No key needed · sample committees replay in demo mode
Research, finance engine, agent personas, 3D committee and this site. Built for work where engineering, capital decisions and consulting meet.