amith-ganta/tessera-multi-tenant-rag

How amith-ganta/tessera-multi-tenant-rag works

Tessera is a multi-tenant document question-answering API. Users authenticate, upload tenant documents for indexing, then ask questions that are routed through retrieval and answer generation; optional evaluation can run synchronously or through a judge queue. The map also includes documented A2A agents, tenant safeguards, streaming, and the quality gate. Request and subsystem links follow the README; sampled code confirms selected API, indexing, routing, and worker behavior, but not every internal call.

Main entry points: app.py FastAPI · orchestrator.py RAG Orchestrator · a2a_supervisor.py A2A Supervisor · redis_queue.py Judge Queue

Components 32

Access And API

Streamlit UI src/ui/app_streamlit_auth.py
FastAPI src/api/app.py
User Authentication src/auth/auth.py
A2A Service Auth src/auth/service_auth.py
Tenant Governance src/resilience/tenant_governance.py
Rate Limiting src/resilience/rate_limiter.py

Knowledge And Answers

Document Ingestion src/rag/ingest.py
Tenant Vector Index src/rag/retriever_dense.py
RAG Orchestrator src/rag/orchestrator.py
Query Router src/rag/router.py
Semantic Cache src/cache/semantic_cache.py
Hybrid Retrieval src/rag/retriever_hybrid.py
Dense Retrieval src/rag/retriever_dense.py
Sparse Retrieval src/rag/retriever_sparse.py
Reranking src/rag/reranker.py
Answer Generation src/rag/generator.py

Evaluation And Agents

Answer Guard src/rag/answer_guard.py
Evaluation Gate evals/gate.py
Judge Queue src/judge/redis_queue.py
Judge Worker judge_worker.py
Judge Results src/judge/judge_store.py
A2A Supervisor src/orchestrator/a2a_supervisor.py
Drafter Agent src/agents/drafter_agent.py
Judge Agent src/agents/judge_agent.py
Agent Checkpoint src/state/redis_checkpointer.py

Operations And State

Redis
RAG Quality Signals src/observability/rag_signals.py
Analytics Log src/rag/analytics.py
User Database src/auth/auth.py

Other components

User / Client
LLM Provider
Tavily Search

Connections 36

User / Client → Streamlit UI: uses
Streamlit UI → FastAPI: sends requests
User / Client → FastAPI: queries / uploads
FastAPI → User Authentication: authenticates
User Authentication → User Database: reads / writes
FastAPI → Rate Limiting: checks limits
FastAPI → Tenant Governance: checks tenant limits
FastAPI → Document Ingestion: indexes uploads
Document Ingestion → Tenant Vector Index: builds index
FastAPI → RAG Orchestrator: runs strategy
RAG Orchestrator → Query Router: routes query
Query Router → Tavily Search: searches current info
RAG Orchestrator → Semantic Cache: checks cache
RAG Orchestrator → Hybrid Retrieval: retrieves context
Hybrid Retrieval → Dense Retrieval: retrieves vectors
Hybrid Retrieval → Sparse Retrieval: retrieves terms
Dense Retrieval → Tenant Vector Index: queries
RAG Orchestrator → Reranking: reranks results
RAG Orchestrator → Answer Generation: generates answer
Answer Generation → LLM Provider: requests completion
FastAPI → Answer Guard: guards answer
Answer Guard → LLM Provider: evaluates / refines
FastAPI → Judge Queue: enqueues evaluation
Judge Queue → Redis: stores jobs
Judge Worker → Judge Queue: claims jobs
Judge Worker → LLM Provider: evaluates answer
Judge Worker → Judge Queue: stores result
FastAPI → Judge Results: reads evaluation
FastAPI → A2A Supervisor: dispatches A2A
A2A Supervisor → A2A Service Auth: authenticates calls
A2A Supervisor → Drafter Agent: requests draft
A2A Supervisor → Judge Agent: requests judgment
A2A Supervisor → Agent Checkpoint: persists state
FastAPI → Evaluation Gate: runs quality gate
FastAPI → RAG Quality Signals: records metrics
RAG Quality Signals → Analytics Log: writes signals

More Python diagrams: jeffser/alpaca · joeynyc/hermes-hudui · aifengheshu/plug-play-modules · pyca/bcrypt · tencent-hunyuan/auk

Drawn by GitDiagram on September 25, 2026 from the repository's file tree, README and source files. Read as Markdown ·