Skip to content

ASK Knowledge Management Service — Overview

The Knowledge Management Service (KMS) is the document intelligence layer of the ASK platform. It handles everything from file upload to RAG retrieval: ingesting documents, chunking and embedding them, storing vectors in Weaviate, and answering retrieval queries from the assistant.


What it does

Responsibility Detail
Document ingestion Accepts file uploads, parses content (PDF, DOCX, images via VLM), chunks text, generates embeddings, and stores them in Weaviate
Vector search Receives retrieval queries from the assistant, performs similarity search against Weaviate, re-ranks results with the Nemotron reranker
Live ingestion Processes real-time document feeds (e.g. web pages, live uploads) through a dedicated worker
Organisation isolation Each organisation's documents are stored in a separate Weaviate collection scoped by org_id
S3 storage Stores raw document files in Ceph RGW (bucket: ask-knowledge-management) and presigns download URLs

Workload architecture

Deployment Replicas Role
knowledge-management-service 1 API — receives upload/query requests, dispatches ingestion jobs
knowledge-management-service-ingestion-worker 1 Batch ingestion — heavy embedding + VLM calls
knowledge-management-service-live-ingestion-worker 1 Real-time ingestion for streaming document feeds
knowledge-management-service-generic-worker 1 Background tasks dispatched via Hatchet

Integrations

System Role
Weaviate Vector store — stores and retrieves 1024-dim embeddings (model: qwen/qwen3-embedding-0-6b)
Foundry LiteLLM Embedding model (qwen/qwen3-embedding-0-6b), VLM for image/PDF understanding (qwen/qwen3-5-122b-a10b-kms), reranker (nvidia/llama-nemotron-rerank-1b-v2)
CNPG knowledge-pg Metadata database — schema knowledge_service. Accessed via knowledge-pg-rw.ask.svc.cluster.local
Ceph RGW (S3) Raw document storage. Endpoint: https://s3.cl1.sq4.aegis.internal, bucket: ask-knowledge-management
Core Service IAM — validates user tokens and resolves org context via api.ask.mod.auh1.dev.dir/ask71/v2/svc
Hatchet Async job queue for ingestion pipelines
Vault / ESO All secrets synced into knowledge-management-service-secret from Vault KV v2

Pages in this section

Page Purpose
KMS Deployment End-to-end deploy runbook — image mirroring, Weaviate vendoring, Helm values, secrets
KMS Config & Model References TOML config walkthrough, all model references, Weaviate apikey wiring
KMS Architecture, Integrations & Ops Weaviate PSA fix, memberlist fix, retriever reranker, day-2 ops
KMS S3 Wiring (Ceph RGW) Bucket creation, IAM policy, and S3 endpoint configuration for document storage