Deep Alpha Copilot
Research  ·  Method  ·  Architecture

Deterministic scores. Agent orchestration. One answer.

Two inputs feed every output. Deterministic analytics — scoring, benchmarks, heuristics — for consistency. Live evidence retrieval — news, filings, sentiment — for context. A synthesis step turns the two into a single, decision-ready response.

Outcome
Decision-ready summary.

Thesis, risks, what to monitor next.

Method
Hybrid reasoning.

Scores, evidence, and synthesis in one flow.

System
Multi-agent orchestration.

A router delegates to specialist tools.

Design
Low-latency runtime.

Fast cache reads; durable cloud storage.

From question to answer

Question, route, tools, answer.

One request. Many internal tool calls. One synthesized response. Consistent across tickers, context-aware to the hour.

Question user prompt Router intent → routing Scoring News & sentiment Filings & flows Answer thesis · risks
Why agents

Break complex requests into smaller tools and pull evidence from the right tools.

Why tool-first

Structured outputs keep conclusions grounded and comparable across tickers.

What you receive

A concise thesis, key risks, and the signals to monitor next.

Hybrid reasoning

Two streams, one synthesis.

Deterministic analytics — scores, benchmarks, heuristics — and evidence retrieval — news, filings, sentiment — feed a synthesis step that produces a single, decision-ready output: thesis, risks, recommendation band, and a monitoring plan.

The structure is deliberate. Scores tell you what is comparable; evidence tells you what is changing. Only synthesis tells you what matters and why, right now.

  • Consistency

    The scoring framework keeps comparisons stable across tickers and time.

  • Context

    Evidence — news and filings — updates the narrative and highlights what changed.

  • Clarity

    Outputs are structured to support decisions, not merely to summarize data.

Agent system

Root, then specialists.

A single root agent coordinates specialist workers behind the scenes. Hierarchical routing. Tool-first. Graceful degradation when a source is slow.

ROOT Financial Root Agent intent · routing · synthesis Company Data scores · facts · rec News headlines · impact Document RAG filings · evidence Sentiment social signals Flows institutional context

Hierarchical routing. Tool-first. Graceful degradation.

› Technical notes

The root is a Gemini-backed agent with a tool registry. Each specialist exposes a structured tool surface so the synthesis layer can read evidence directly without re-parsing prose.

Specialists share a common cache layer. Hot-path queries hit a local cache (sub-second) before falling through to durable storage (a few seconds on a cold ticker).

If a specialist times out or returns malformed evidence, the root proceeds with what it has and labels the gap explicitly in the synthesis.

Production design

Fast read, durable store.

The production topology separates the user path — low-latency, cached — from the durable store and the analytics warehouse. Requests never wait on slow data; reports never slow production.

  • Hot path
    Fast user path
    /tmp · /cdn

    Real-time queries read from the local cache. Sub-second for cached tickers, a few seconds on cold path.

  • Durable
    Durable foundation
    Cloud Storage

    Artifacts persist across restarts and deployments. Nothing is lost between warm boots.

  • Analytics
    Analytics layer
    BigQuery

    Warehouse supports reporting and backtests without slowing production reads.

Where to go next

The fastest way to understand the methodology is to run it on a name you already follow.

Browse the coverage → About the product Methodology

Disclaimer. Deep Alpha Copilot provides informational analysis and does not constitute investment advice.