Equities RAG is the next specialized stack inside Hermes. Same shape as the Polymarket stack — its own signal layer, its own quant layer, its own execution path — but tuned for stocks, ETFs, and indices rather than probability markets. Trading Hermes already trades equities through its general path; Equities is what makes that trading good, the same role the Polymarket stack plays for prediction markets.
The problem I'm wrestling with isn't the shape. The shape carries over. The problem is the data.
Polymarket has narrow data. A specific market resolves to a specific outcome; the signal universe around it is contained. Equities is the opposite. Pick any ETF — SPY, the sector funds, anything broad — and the relevant context isn't a document or a feed. It's the constituents, the sector flow, the macro regime, the fund flows, the options chains, the earnings cycle of every name inside, the news everywhere those names trade. The industry sprawl is the problem.
When I query the agent about a specific ETF, retrieval pulls broad industry context instead of the specific entity. The vector DB is doing what it was told to do — surface relevant material — but relevant at industry scale is half the index. The agent loses focus before it produces signal.
I haven't fixed this yet. The cleanest path I'm considering shards retrieval by entity tier — index, constituent, sector, macro — and routes the query through tiers rather than across them. That's the idea. Whether it works is a build away.