How entbappy/atlas-enterprise-ai works
Atlas is a document-grounded knowledge assistant. Users upload organizational files, the API processes and indexes them, and chat requests pass input checks, retrieve authorized evidence, generate and validate an answer, and persist conversation turns. The dashboard also exposes document status, approvals, feedback, and operational metrics. The graph includes documented integrations and sampled call paths; some UI routes and storage details are not fully shown in the excerpts.
Main entry points: main.py FastAPI routes · graph.py RAG workflow · ingest.py Parse and chunk · vector.py Authorized retrieval
Components 23
Web experience
- Web dashboard
app/templates/index.html - Browser interactions
app/static/js/app.js - FastAPI routes
app/main.py - Demo principal
app/demo_context.py
Document ingestion
- Upload completion
app/main.py - Parse and chunk
worker/ingest.py - Embedding client
app/aws.py - Amazon S3
Knowledge assistant
- Pinecone index
app/vector.py - RAG workflow
app/graph.py - Input guardrails
app/guardrails.py - Authorized retrieval
app/vector.py - Grounded generation
app/aws.py - Output guardrails
app/guardrails.py - Chat adapter
app/rag.py - Conversation memory
app/memory.py
Governance and operations
- PostgreSQL
- Approvals and feedback
app/main.py - Audit events
app/db.py - Rius tracing
app/observability.py
Other components
- User / browser
- Amazon Bedrock
- Workspace administrator
Connections 32
- User / browser → Web dashboard: uses
README.md - Web dashboard → Browser interactions: loads
README.md - Browser interactions → FastAPI routes: calls
README.md - FastAPI routes → Demo principal: resolves principal
app/main.py - FastAPI routes → Amazon S3: presigns upload
app/main.py - User / browser → Amazon S3: uploads file
README.md - FastAPI routes → Upload completion: accepts completion
app/main.py - Upload completion → Parse and chunk: calls
app/main.py - Parse and chunk → Amazon S3: reads file
worker/ingest.py - Parse and chunk → Embedding client: embeds chunks
worker/ingest.py - Embedding client → Amazon Bedrock: invokes model
app/aws.py - Parse and chunk → Pinecone index: upserts vectors
worker/ingest.py - Parse and chunk → PostgreSQL: updates status
worker/ingest.py - FastAPI routes → Chat adapter: calls
app/main.py - Chat adapter → RAG workflow: delegates to
app/rag.py - RAG workflow → Input guardrails: checks input
app/graph.py - RAG workflow → Authorized retrieval: searches
app/graph.py - Authorized retrieval → Embedding client: embeds query
app/graph.py - Authorized retrieval → Pinecone index: queries
app/vector.py - RAG workflow → Grounded generation: generates answer
app/graph.py - Grounded generation → Amazon Bedrock: converses
app/aws.py - RAG workflow → Output guardrails: checks output
app/graph.py - FastAPI routes → Conversation memory: loads and saves turns
app/main.py - Conversation memory → PostgreSQL: reads and writes
app/memory.py - FastAPI routes → PostgreSQL: stores documents
app/main.py - FastAPI routes → Approvals and feedback: serves review and feedback
app/main.py - Workspace administrator → FastAPI routes: reviews approvals
app/main.py - Approvals and feedback → PostgreSQL: reads and writes
app/main.py - FastAPI routes → Audit events: records events
app/main.py - Audit events → PostgreSQL: inserts events
app/db.py - FastAPI routes → Rius tracing: initializes tracing
app/main.py - RAG workflow → Rius tracing: decorated for traces
app/graph.py