How kasturimaheshbabu2316-mc/capstone---bank works
Cred Domain Support Agent is an offline lending-support service. A customer submits a policy or application-status question through the API; request controls lead into a multi-agent workflow, with policy retrieval or application lookup, structured response generation, and secondary compliance review. The repository also documents document ingestion, session memory, caching, budgeting, and audit logging. The README and excerpts establish the main workflow, but do not expose every integration detail, so unsampled subsystem links are limited to documented relationships.
Main entry points: app.py HTTP and WebSocket API · main.py FastAPI gateway · crew.py Support agent crew · vector_store.py Local vector retrieval
Components 20
Service interface
- HTTP and WebSocket API
api/app.py - FastAPI gateway
app/main.py - Response contracts
app/models.py - Request audit logger
api/logger.py
Agent workflow
- Pipeline coordinator
app/pipeline.py - Support agent crew
agents/crew.py - Deterministic LLM
agents/mock_llm.py - Agent response schema
agents/schemas.py - Session memory
agents/memory.py - Compliance peer review
review/autogen_review.py
Knowledge and records
- Lookup and escalation tools
agents/tools.py - Seeded applications
dataset.py - Banking policy documents
knowledge_base - Policy chunking
rag/chunking.py - Local vector retrieval
rag/vector_store.py - Application lookup store
app/db.py
Safety and governance
- Input and output guardrails
agents/guardrails.py - Request budget guard
governance/budget_guard.py - Query response cache
governance/cache.py
Other components
- Customer
Connections 30
- Customer → HTTP and WebSocket API: submits inquiry
README.md - Customer → FastAPI gateway: submits inquiry
README.md - HTTP and WebSocket API → Request budget guard: checks budget
api/app.py - HTTP and WebSocket API → Input and output guardrails: guards input
api/app.py - HTTP and WebSocket API → Query response cache: checks cache
api/app.py - HTTP and WebSocket API → Support agent crew: runs workflow
api/app.py - HTTP and WebSocket API → Compliance peer review: reviews draft
api/app.py - HTTP and WebSocket API → Request audit logger: logs request
api/app.py - HTTP and WebSocket API → Policy chunking: chunks documents
api/app.py - HTTP and WebSocket API → Local vector retrieval: indexes documents
api/app.py - FastAPI gateway → Pipeline coordinator: runs workflow
app/main.py - FastAPI gateway → Request budget guard: checks budget
app/main.py - FastAPI gateway → Input and output guardrails: guards input
app/main.py - FastAPI gateway → Query response cache: checks cache
app/main.py - FastAPI gateway → Request audit logger: logs request
app/main.py - FastAPI gateway → Local vector retrieval: checks grounding
app/main.py - Pipeline coordinator → Support agent crew: runs primary crew
app/pipeline.py - Pipeline coordinator → Compliance peer review: reviews draft
app/pipeline.py - Support agent crew → Deterministic LLM: uses inference
agents/crew.py - Support agent crew → Session memory: reads and records
agents/crew.py - Support agent crew → Lookup and escalation tools: calls lookup
agents/crew.py - Support agent crew → Agent response schema: builds response
agents/crew.py - Compliance peer review → Agent response schema: reviews response
review/autogen_review.py - Lookup and escalation tools → Seeded applications: looks up records
agents/tools.py - Banking policy documents → Policy chunking: chunks policies
README.md - Policy chunking → Local vector retrieval: indexes chunks
rag/vector_store.py - Support agent crew → Local vector retrieval: retrieves policy
- Application lookup store → Seeded applications: loads records
app/db.py - HTTP and WebSocket API → Response contracts: returns structured response
api/app.py - FastAPI gateway → Response contracts: returns structured response
app/main.py