gargi-09/museum-qa-agent

How gargi-09/museum-qa-agent works

Museum Records QA ingests museum-catalogue data, then answers natural-language questions with hybrid local retrieval and one bounded model call. The query workflow enriches retrieved records with curator corrections, duplicate/entity grouping, and contradiction checks before deterministic verification and confidence scoring; results retain cited, used, and unused records. The README documents persistent normalized data, embedding-cache invalidation, and correction overrides. Edges reflect the excerpts and README; unsampled implementation details are not inferred beyond documented capabilities.

Main entry points: reasoning.py Answer orchestration · main.py CLI commands · retrieval.py Hybrid retrieval · ingest.py Record normalization

Components 15

Corpus preparation

Record normalization src/ingest.py
Normalized corpus

Question answering

CLI commands src/main.py
Hybrid retrieval src/retrieval.py
Embedding cache
Answer orchestration src/reasoning.py
Verified answer src/reasoning.py

Record quality

Entity resolution src/entity_resolution.py
Prose contradiction checks src/contradiction.py

State and services

Correction persistence src/memory.py
Correction records
Model API client src/api_client.py

Other components

Curator
Questioner
Model endpoint

Connections 18

Curator → CLI commands: adds corrections README.md
Questioner → CLI commands: asks question README.md
CLI commands → Record normalization: runs ingest src/main.py
Record normalization → Model API client: fallback extraction src/ingest.py
Record normalization → Normalized corpus: writes records README.md
CLI commands → Hybrid retrieval: loads for queries src/main.py
Normalized corpus → Hybrid retrieval: loads corpus README.md
CLI commands → Answer orchestration: dispatches question src/main.py
Hybrid retrieval → Embedding cache: caches embeddings README.md
Answer orchestration → Hybrid retrieval: ranks candidates src/reasoning.py
Answer orchestration → Correction persistence: applies overrides README.md
Correction persistence → Correction records: reads and writes src/memory.py
Answer orchestration → Entity resolution: groups records src/reasoning.py
Answer orchestration → Prose contradiction checks: checks descriptions src/reasoning.py
Answer orchestration → Model API client: requests generation src/reasoning.py
Model API client → Model endpoint: sends model request README.md
Answer orchestration → Verified answer: verifies and scores src/reasoning.py
CLI commands → Verified answer: prints result src/main.py

More Python diagrams: aws-samples/eks-cluster-upgrade · aws-solutions/mlops-workload-orchestrator · cognizant-ai-lab/neuro-san · facebookresearch/neural-rewriter · impredicative/podgenai

Drawn by GitDiagram on October 13, 2026 from the repository's file tree, README and source files. Read as Markdown ·