Industrial Hypertext · Perth, Western Australia
RAG chatbots and document assistants
Help staff find answers in your own documents, procedures and operational knowledge. We build retrieval pipelines and the applications around them, or configure a self-hosted search platform where that fits better.
Discuss a document assistantIndustrial Hypertext · Software development since 2002
Make documents easier to use
Teams in mining, oil and gas, healthcare and other document-heavy environments often spend time searching folders, PDFs and spreadsheets. A retrieval-augmented generation (RAG) assistant retrieves relevant source material and uses it to help answer a question.
The application can show references so staff can check the answer against the original material. We evaluate retrieval and answer quality using real questions from your team, including cases where the documents do not contain an answer.
Build the retrieval around your material
We work on document ingestion, chunking, embeddings, indexing and the chat interface. The pipeline needs to handle your actual formats and the way the material is updated, including scanned documents where OCR is required.
Custom implementations can use PostgreSQL and pgvector. MineAction, our mining document assistant, provides a working example of this approach. Access rules and source references are part of the application design.
Choose a custom build or a self-hosted platform
A fully custom application is useful when the interface or workflow needs specific behaviour. A self-hosted Onyx deployment is another option for teams that want an existing enterprise search and chat platform configured for their environment.
We assess connectors, permissions, model hosting and any custom ingestion work before choosing the approach. Where sensitive material must stay within your environment, that requirement guides both retrieval and model configuration.
Start with a representative document collection
The first pilot should include the material people actually need to search, rather than only a few easy examples. We agree the sources, users, questions and acceptance criteria before expanding the corpus.
Larger estates can be introduced in stages, with ingestion monitoring, deduplication and format handling scoped as the collection grows. Costs and hosting requirements depend on those choices.
Connect answers to the wider workflow
Some assistants need document retrieval and live business data. A custom MCP integration can expose approved system queries alongside the document assistant.
See AI app development and integrations for broader projects. Tell us where your material lives and what questions your staff struggle to answer today.
Start with a conversation
Tell us what you would like built or improved. A few sentences are enough to discuss fit, scope and the next step. Response target: one business day.
Discuss a document assistant