How nishu2000-hub/consultiq-agent works
ConsultIQ accepts a business question in a Streamlit interface, runs an autonomous research agent that can search the web and retrieve public-company financials, and presents a structured consulting memo. A separate Claude-based judge evaluates the memo across five dimensions, and the interface displays scores and a PASS/REVIEW/FAIL decision. The agent loop and evaluator are supported by sampled source; some UI invocation details are omitted, so those links rely on the README’s workflow description.
Main entry points: agent.py Agent Reasoning · evaluator.py Memo Judge · app.py Streamlit Interface
Components 14
User Experience
- Streamlit Interface
app.py - Strategy Memo
agent.py
Research Agent
- Agent Reasoning
agent.py - Tool Router
agent.py - Tool Runner
agent.py - Claude Agent Model
Research Sources
- Web Search Tool
agent.py - Financial Data Tool
agent.py - Tavily Search
- yfinance
Memo Evaluation
- Memo Judge
evaluator.py - Claude Judge Model
- Quality Gate
app.py
Other components
- Business User
Connections 15
- Business User → Streamlit Interface: enters question
README.md - Streamlit Interface → Agent Reasoning: submits question
README.md - Agent Reasoning → Claude Agent Model: reasons with
agent.py - Agent Reasoning → Tool Router: checks tool calls
agent.py - Tool Router → Tool Runner: dispatches tools
agent.py - Tool Runner → Web Search Tool: runs
agent.py - Tool Runner → Financial Data Tool: runs
agent.py - Web Search Tool → Tavily Search: searches with
agent.py - Financial Data Tool → yfinance: retrieves data with
agent.py - Tool Runner → Agent Reasoning: returns tool results
agent.py - Agent Reasoning → Strategy Memo: synthesizes
README.md - Strategy Memo → Memo Judge: evaluated by
README.md - Memo Judge → Claude Judge Model: scores with
evaluator.py - Memo Judge → Quality Gate: returns evaluation
README.md - Quality Gate → Streamlit Interface: displays verdict
README.md