Case Study

SWFI — Institutional Capital Intelligence

Real-time, evidence-gated AI search across the global institutional investor landscape.

The Challenge

Sovereign wealth funds, public pensions, central banks, and endowments operate within a fragmented intelligence landscape. The Sovereign Wealth Fund Institute (SWFI) — the leading data provider for this market — maintains profiles on more than 27,000 institutional investors, tracking transactions, mandates, and personnel changes in real time.

Decision-makers at these institutions do not need another search bar that returns raw tables. They need synthesized, evidence-backed briefs that resolve ambiguous queries against verified records and produce answers they can act on with confidence.

The Approach

Active Mirror built a governed AI intelligence layer on top of SWFI's institutional dataset, deployed at institutionalinvestorhub.com. The core constraint: every response must trace to identified source records. If the system cannot find evidence to support a claim, it states that explicitly rather than generating unsourced analysis.

A graph-based entity resolution engine classifies user intent across 10 distinct query types — rankings, comparisons, mandate searches, sector scans, people lookups, transaction histories, trend analysis, and more — then assembles context from the appropriate data sources before generating a response. Access is tiered between public and subscriber-gated routes.

The Architecture

The query router classifies intent, resolves entities against the institutional graph, and enforces evidence-class rules at the response layer. A sanitizer inspects every generated response and replaces any claim that cannot be traced to a verified record. Security includes rate limiting, prompt injection detection, origin verification, and X-Robots-Tag enforcement on sensitive endpoints. A status API masks underlying data from unauthenticated requests.

The Results

27,249
Institutions
177,677
Transactions
128,781
People Records
4,800+
Mandates

"Evidence over eloquence. The system produces analyst briefs, not chatbot answers."