Skip to content
Case study

Private AI knowledge assistant for a logistics operator

A private, on-premise WislaSearch deployment that turns a logistics operator's scattered operational knowledge into instant, cited answers - without anything leaving their network.

Industry
Artificial Intelligence, Logistics
Tech stack
RAG, Qdrant, MongoDB, LoRA fine-tuning, hybrid search
OutcomeWhat the work delivered
1.00
answer faithfulness
0.99
answer correctness
0.98
answer relevancy
0.96
context relevance
The challengeThe brief, and what was at stake

A large international logistics and terminal operator runs on deep operational knowledge - procedures, safety instructions and system manuals - spread across documents and internal systems. Staff lost time hunting for the right instruction, a handful of experts spent their days answering the same questions, and onboarding a new hire was slow. They needed faster answers without sending sensitive operational data outside their own network.

Have a similar challenge?Have an integration or build blocking revenue?

Bring one stuck or upcoming project - we will diagnose it with you and scope the fastest safe path to live.

How we built itWhat WislaCode designed and shipped
01Indexed their knowledge, kept it private

We deployed WislaSearch, WislaCode's on-premise AI knowledge assistant, entirely inside the client's environment. It indexes their internal documents into a hybrid store - semantic, keyword and vector search over a Qdrant vector index with a MongoDB document store - so nothing ever leaves the client's infrastructure and no external model sees their data.

The WislaSearch assistant running on a laptop, indexing a private knowledge base
02Answers grounded in the source

Staff ask questions in plain language and get a direct, context-aware answer that cites the source document. The model was fine-tuned to the client's domain language (LoRA), the production model is configured to the client's security posture (local or cloud), and access is governed through their corporate directory, so each person sees only what they are cleared to. On the deployment's own evaluation the answers scored 0.98 for relevancy, 0.96 for context relevance, 0.99 for correctness and 1.00 for faithfulness; in daily use, support tickets fell, instruction lookup got far faster, the load on subject-matter experts eased, and new hires got up to speed sooner.

WislaSearch in three steps - upload sources, build the knowledge base, ask and get cited answers
This is a WislaCode productWislaSearch is a product you can pilot

This deployment runs on WislaSearch - WislaCode's own on-premise AI knowledge assistant. The same product is available to pilot in about a week.

Have a similar challenge?Let's map the fastest safe path to live

Book a 30-minute integration review. Bring one stuck or upcoming integration - we'll diagnose it with you and scope the work, no obligation.