RAG
Finds the evidence you need right now.
Finds the relevant passages and where they are, in sources the user is allowed to see.
- Legislation, case law and patent texts
- Manuals and work policies
- Case, customer and project files
Evidence from your own material,
answers by your own standards.
Evidence comes from equipment data, regulations, patents and work documents, found when it's needed. How you write and review is what AI learns.
Does every AI answer
still need someone
to rewrite it?
Feeding it documents doesn't pass on how you work.
The latest documentsFind them again
In your organization's formatWrite it again
Missing items and wordingPolish it again
Prepare your work materials by separating
documents to search and writing standards to learn.
For AI to learn your work,
what do you prepare first?
Find evidence in work documents and learn writing standards from reviewed examples. Use RAG for retrieval and fine-tuning for training.
Material
Invention material
Invention disclosures, specifications, drawings and claims
RAG
Current patent information
KIPRIS candidates, publication and registration details, and prior art texts
Fine-Tuning
Your comparison format
Claim element breakdown and review table structure
For review
A comparison draft linked to evidence
Documents, elements and follow-up searches for experts to check
Patent attorneys and the experts in charge judge novelty, inventive step and infringement, and handle submission.
Use RAG for retrieval
and fine-tuning for your writing standards.
Use retrieval of current source material and training on document-writing practices as distinct parts of one workflow.
Material and methodmeet in one draft→
Fine-Tuning
Learns your organization's standards.
From reviewed examples and revision history, the model learns the structures and phrasing you use most.
- Document structure and style
- Categories and status values
- Review order and completeness checks
Carries through to review and approval.
The draft is reviewed and approved by the person in charge, then saved.
- Collect material
- Find evidence
- Your format
- Expert review
- Approve and save
Indexing and training,
both visible on screen.
See how far RAG indexing has got and how training data is being prepared, from the admin console.
RAG
See which material is searchable.
New material only needs indexing, not training, to show up in search.
Fine-Tuning Plugin
Turns work files into training data.
Converts XML, HTML, JSON and TXT sources to Markdown and builds PT or SFT datasets.
Does your organization
need Fine-Tuning?
Start by retrieving the sources you need with RAG. Consider fine-tuning when reviewed results accumulate and the same writing or classification standards need to be applied repeatedly.
Start with RAG when
- Material changes often
- Originals and sources matter
- Each question needs different facts
Look at Fine-Tuning when
- Validated examples keep repeating
- You have a fixed structure and classification rules
- The same errors keep getting corrected
Work that requires expert judgment
- Final legal and medical judgments
- New exceptions without enough evidence
- Approvals and actions that carry responsibility
For each industry,
define the sources to search and standards to learn.
Choose source material and document formats for the industry. The responsible expert makes the final decisions.
| Industry | Found with RAG | Learned with Fine-Tuning | Decided by people |
|---|---|---|---|
Legal and patentsProfessional documents with source evidence to check | Current legislation, case law, patent texts and case files | Pleading structure, claim comparison tables and review style | Legal strategy, patentability and final submission |
Manufacturing and equipmentFrom field records to reports you can act on | Manuals, inspection history, error codes and equipment documents | Anomaly categories, action reports and field terminology | Confirming the fault, deciding on maintenance and operation |
Healthcare and researchSource records and study formats kept together | Source records, protocols, study documents and supporting papers | CRF narratives, classification criteria and review templates | Medical judgment, participant safety and study approval |
Finance and insuranceRegulations and your own review criteria applied together | Product terms, regulations and customer submissions | Review items, statements of reasons and risk classification formats | Approval, rejection, exceptions and risk decisions |
Education and contentResults using current sources and your writing standards | Textbooks, course material and current references | Feedback style, rubrics and content structure | Learning goals, assessment and publishing decisions |
Before training,
decide what a good result looks like.
- Pick a typical task
Pick one task that repeats and whose results can be judged.
- Sort data from criteria
Split the current knowledge to search from the good examples to learn.
- Measure before training
Record the model's errors and experts' correction time.
- Train and test small
Compare format, omissions and evidence on the same evaluation set.
- Rollout and refinement
Apply it where experts approve, and keep collecting fixes.
Start with one repeated task
and assess how AI fits your organization.
Use work documents and reviewed examples to define what to retrieve, which writing standards to train and where staff approval is required.