mugnihidayah/synapse-instant-document-insight

How mugnihidayah/synapse-instant-document-insight works

Synapse is a FastAPI multimodal document-Q&A service. Its main workflow is API-key-authenticated session creation and upload, optional background ingestion, extraction and chunk storage, then grounded answers with citations; users can also inspect sessions, export history, submit feedback, and review usage. The graph preserves README-documented capabilities such as OCR, hybrid retrieval, agentic tools, and query filters. Some subsystem wiring is documented rather than shown in the sampled source, so uncertain call relationships are omitted.

Main entry points: query.py Query routes · main.py FastAPI application · documents.py Document routes · hybrid_search.py Hybrid search

Components 30

API and sessions

FastAPI application src/api/main.py
API key routes src/api/routes/keys.py
Document routes src/api/routes/documents.py
Query routes src/api/routes/query.py
Insights routes src/api/routes/insights.py
Session service src/api/session.py
API key auth src/api/auth.py

Document ingestion

Async ingestion jobs src/api/ingestion_jobs.py
Ingestion contract src/api/ingestion_contract.py
File loaders src/ingestion/loaders.py
Document chunking src/ingestion/chunkers.py
Metadata filters src/ingestion/metadata_filters.py
Original file storage src/ingestion/file_storage.py

RAG and agents

Query orchestration src/api/routes/query.py
History contextualization src/rag/contextualize.py
Query rewriting src/rag/query_rewrite.py
Hybrid search src/rag/hybrid_search.py
Result reranking src/rag/reranker.py
Retrieval tuning src/rag/retrieval_utils.py
Grounding checks src/rag/grounding.py
Answer generation src/rag/chain.py
Agent orchestration src/agent/orchestrator.py
Agent tools src/agent/tools.py
Chat history src/rag/chat_history.py

Persistence and usage

Vector storage src/ingestion/pgvector_store.py
PostgreSQL and pgvector src/db/connection.py
Database models src/db/models.py
Usage and quota src/api/usage.py

Other components

API client
LLM and embedding providers

Connections 36

API client → FastAPI application: sends requests
FastAPI application → Document routes: dispatches
FastAPI application → Query routes: dispatches
FastAPI application → API key routes: dispatches
FastAPI application → Insights routes: dispatches
Document routes → API key auth: authenticates
Query routes → API key auth: authenticates
Insights routes → API key auth: authenticates
Document routes → Session service: manages sessions
Document routes → Original file storage: persists uploads
Document routes → Async ingestion jobs: schedules ingestion
Async ingestion jobs → Session service: updates status
Async ingestion jobs → Ingestion contract: processes payloads
Ingestion contract → File loaders: extracts content
Ingestion contract → Document chunking: creates chunks
Ingestion contract → Vector storage: stores chunks
File loaders → LLM and embedding providers: uses OCR
Vector storage → PostgreSQL and pgvector: reads and writes
API key routes → PostgreSQL and pgvector: stores key records
Session service → PostgreSQL and pgvector: reads and writes
Query routes → Session service: loads session
Query routes → Chat history: loads chat history
Query routes → History contextualization: contextualizes query
Query routes → Hybrid search: retrieves documents
Hybrid search → Vector storage: vector search
Hybrid search → PostgreSQL and pgvector: keyword search
Hybrid search → Metadata filters: applies filters
Query routes → Result reranking: reranks results
Query routes → Answer generation: generates answer
Query routes → Grounding checks: checks grounding
Query routes → Agent orchestration: runs agent mode
Agent orchestration → Agent tools: executes tools
Agent orchestration → Hybrid search: retrieves sources
Answer generation → LLM and embedding providers: requests completion
Insights routes → Usage and quota: reports usage
Usage and quota → PostgreSQL and pgvector: reads and writes

More Python diagrams: cloudera/cdpcli · cmach508/paflow · coder-irwin/voice-agent-workshop · computersciencehouse/conditional · crate/cratedb-toolkit

Drawn by GitDiagram on September 28, 2026 from the repository's file tree, README and source files. Read as Markdown ·