Engineering in action

Look inside the system.

Explore the decisions behind the interface: where context comes from, how tools are used, and where human judgment stays in control.

Interactive walkthrough · Synthetic data

From customer question
to controlled action.

Follow a representative support request through retrieval, business tools, and human review. This simulation illustrates the documented design; it does not call a live AI service.

  1. 01
    Understand the requestCustomer → Agent
  2. 02
    Retrieve relevant contextRAG → Policy context
  3. 03
    Inspect the business resultC#/.NET MCP → Order tools
  4. 04
    Respond with the available evidenceAgent → Customer

Step 1 / 4

Understand the request

The customer asks where an order is. The workflow needs order context rather than a plausible shipping estimate.

Illustrative request

“Where is my order DEMO-1042?”

Engineering decision

Distinguish a factual lookup from a general policy question before choosing a business tool.

All names, order numbers, policies, and tool outputs above are invented examples. The client implementation and its reported outcome are described in the case study →.

Architecture explorer · NovaKnowledge

Inspect the boundaries.
Understand the tradeoffs.

Select a component to see its responsibility, failure considerations, and implementation evidence. This is a source-backed architecture explanation, not a running deployment.

Ingest → Retrieve → Orchestrate → Answer
Evaluation and operations support the pipeline.

Component 01

Document ingestion

Document processing extracts text, creates chunks, optionally generates embeddings, and indexes content for retrieval.

Design decision
Process ingestion asynchronously. Keep document status explicit, replace prior chunks, and index in batches.
Failure & scope
Processing errors mark a document Failed and are rethrown. This is not a claim of exactly-once delivery or an automatically recovered pipeline.
Inspect implementation on GitHub ↗

NovaKnowledge is a single-tenant reference implementation without authentication. Hybrid retrieval needs configured embeddings; evaluation reports its engine. Read the full case study →

Discuss a system you're building ↗