rajdeepofgithub/query-router

How rajdeepofgithub/query-router works

Query Router classifies JPMorgan Chase questions as SQL, RAG, or both, then executes the selected backend path and returns an answer; combined answers add disclosures, citations, and conflict handling. Its evaluation workflows measure routing alone and end-to-end answer quality against separate truth data, while shared instrumentation records API usage and cost. The README documents backend stages and integrations that are outside this repository; those are shown as external systems, not as inspected source modules.

Main entry points: router.py Answer entry point · execute.py Route executor · run_router_eval.py End-to-end evaluator · run_routing_eval.py Routing evaluator

Components 14

Routing

Route classifier src/classify.py
Answer entry point src/router.py

Execution

Route executor src/execute.py
Text-to-SQL
Financial RAG
Answer merger src/execute.py

Evaluation

Routing questions data/routing_eval.jsonl
Answer truth data data/router_answer_truth.jsonl
Routing evaluator src/run_routing_eval.py
End-to-end evaluator src/run_router_eval.py
Evaluation results baseline

Observability

API cost tracker src/costs.py

Other components

Question asker
OpenAI API

Connections 17

Question asker → Answer entry point: asks question README.md
Answer entry point → Route classifier: classifies src/router.py
Answer entry point → Route executor: executes route src/router.py
Route executor → Text-to-SQL: runs SQL README.md
Route executor → Financial RAG: runs retrieval README.md
Route executor → Answer merger: merges both src/execute.py
Routing questions → Routing evaluator: supplies questions src/run_routing_eval.py
Routing evaluator → Route classifier: classifies src/run_routing_eval.py
Routing evaluator → Evaluation results: writes results src/run_routing_eval.py
Routing questions → End-to-end evaluator: supplies questions src/run_router_eval.py
Answer truth data → End-to-end evaluator: supplies grading truth src/run_router_eval.py
End-to-end evaluator → Answer entry point: runs answer src/run_router_eval.py
End-to-end evaluator → Route executor: grades outputs src/run_router_eval.py
End-to-end evaluator → Evaluation results: writes results src/run_router_eval.py
Answer entry point → API cost tracker: labels usage src/router.py
API cost tracker → OpenAI API: wraps API calls src/costs.py
Route classifier → OpenAI API: requests classification src/classify.py

More Python diagrams: pineapplek/ed2mol · primeintellect-ai/prime-envs · santanderai/autoguardrails · soupday/ccic-blender-pipeline-plugin · spiderforest/mixercseg

Drawn by GitDiagram on September 29, 2026 from the repository's file tree, README and source files. Read as Markdown ·