Type any geopolitical event or financial scenario. Get a blast radius: every exposed company mapped by supply chain, SEC filings, and historical reaction.
Every time a geopolitical headline breaks, the same question follows: which companies are actually in the blast radius? Most tools stop at the news. This one maps the equity exposure across supply chains, SEC filings, and historical price reactions, and hands you an answer in under a minute.
The engine runs in five stages. A static geography database instantly identifies companies by region and chokepoint with no API call and no latency, now including a curated Oslo Børs ticker set alongside the SEC universe. Before any company is selected, a reasoning step analyzes the event structure: what is the actual transmission mechanism, which sectors have a genuine causal link, and which ones just share a keyword. This filters out the noise that most automated tools include. A database of 198+ companies with scenario and sector tags then handles most lookups instantly, cutting unnecessary AI calls. Revenue exposure is no longer a keyword match: SEC XBRL geographic segment data and Exhibit 21 subsidiary jurisdictions are parsed directly from the filings, and mentions are capped so a passing reference can't outweigh real exposure. An OFAC-backed sanctions screen (SDN and Consolidated lists, fuzzy-matched) flags direct sanctions exposure for free, no API key required. Finally, yfinance pulls how similar events moved these stocks historically, not as prediction, but as pattern context from analogous crises, sharpened by a momentum-confirmation factor that drives the directional call.
The cost structure was engineered deliberately. Different models are used based on task intensity: lightweight classification runs on Haiku at a fraction of a cent, while the intelligence brief that synthesizes everything uses Sonnet. Scenario simulations project sector-level moves using a database of 94 calibrated historical scenarios, scaled by event severity and relationship context. Enrichment runs in parallel, so a full run finishes in seconds, not minutes. A full advanced analysis costs around three cents.
The model is backtested, not just asserted: 266 predictions across 34 historical events, a 66.9% directional hit rate, and a 0.43 rank correlation between predicted and realized moves, each confidence score in the UI is the measured hit rate for that call type, not a guess. Ideas that didn't hold up, like an ex-ante sector factor, are documented as rejected rather than quietly dropped.
I built the first version over two days during my second year at BI Norwegian Business School, where I study finance. I'm now in my third year, 22, and still building on it. I had no prior full-stack experience. Claude was the collaborator throughout, less like using a tool and more like thinking out loud with someone who could immediately turn the idea into working code. The bottleneck shifted from implementation to judgment: what actually matters, what to cut, what to build next.
The result is something I use myself. When a sanctions package drops or a conflict escalates, I open this before I open anything else. If you are a recruiter, investor, or just someone who finds this useful, reach out.