Fine-tuning, RAG or training your own model?
Choose RAG, LoRA fine-tuning or a small custom model for a business task. Understand training data, evaluation and adaptation costs with Syntalith.
custom-modelsllmprivate-aifine-tuning
BLOGCategory
Custom models
Articles grouped by business problem, industry, and system type.
Jev 1.13 for business classification and routing
Where should you test Jev and small decision models? Explore ticket routing, document completeness, confidence thresholds and a Syntalith-led pilot.
custom-modelsllmprivate-aifine-tuning
Local Jev 1.13 alternatives: Kev and TinyJev
Can Jev run locally? Explore Kev, TinyJev and other approaches to private decision models, with selection, fine-tuning and deployment from Syntalith.
custom-modelsllmprivate-aifine-tuning
A model for ranking catalog search results
A customer searches for a product by describing what they need it for, but the first results match only part of the wording. The catalog manager wants a better ranking while retaining control over what actually answers the query. A customized model is one option, but first establish whether the right products reach the candidate list at all.
AI modelsmodel customization
Accepting a customized model for team use
A buyer sees correct answers from a customized model and is told the project is ready. Before accepting it, they need to know whether the assessed version is intended for their application, which errors remain, and who handles cases requiring review. Those agreements should precede the final assessment.
AI modelsmodel customization
Custom or general model: compare the work your team actually does
A customized model may produce a better first draft and still be the wrong purchase for a team. Employees wait for answers, check them, and make corrections, while the company maintains another model version. A useful comparison captures that work in the task the system is actually meant to perform.
AI modelsmodel customization
Custom AI models for rare operational events
A model handles everyday documents well, but the company also wants it to catch occasional exceptions. A few correctly recognized examples leave open the question of how the model handles other wording. For rare events, the assessment needs to cover the available examples and the ordinary cases the model needlessly sends for review.
AI modelsmodel customization
Training material: who approves its use?
A team can open company documents, but planning a model raises another question: which material can be used in this particular project? A shared folder may contain company instructions, customer files, and supplier material. Starting the work requires clarity about where they came from and how they will be used.
AI modelsmodel customization
Business address quality: when is validation enough, and when should you assess AI?
Suppose an order contains a complete customer address, but a note says, “The warehouse receives deliveries.” An employee needs to establish whether the record points to the right location. An AI model can help flag that uncertainty and show the evidence behind it. Before choosing model customization, compare that help with ordinary rules and an address validation service.
business addressesdata qualitymodel customization
Is your company's archive ready for fine-tuning?
Your company has an archive of employee responses and wants to use it to customize an AI model. The number of records looks promising, but it does not tell you whether they demonstrate the work the model needs to do. Before commissioning training, establish which examples are useful, what needs correction, and whether an existing model already handles the task.
fine-tuningdata preparationAI models
Should employee corrections train the next model?
Employees correct AI answers, and the application owner wants to use those edits in the next model update. A “save correction” button collects material but does not explain what the model should learn from it. An edit may fix an error, add a new fact, or meet an individual recipient's request.
AI modelsmodel customization
Fine-tuning or RAG: what will improve product answers?
Suppose a sales assistant describes a product as available even though the order system now marks it unavailable. The team considers further model training. First, establish where the assistant got its information: did it use an old product sheet, or did it receive the current status and misrepresent it? That distinction determines what work is worth commissioning.
product informationRAGfine-tuning
Fine-tuning when the base model provider changes
A company wants to change model providers, but its application uses a version customized for its work. Switching the service alone does not establish whether the new model will behave as expected. The team needs to determine which earlier work remains useful and what requires a fresh assessment.
AI modelsmodel customization
The process changed. Does the model need retraining?
A company changes how cases are handled, but its customized model keeps suggesting the old labels. The manager must decide whether to commission more training, update application rules, or clarify the new process first. Maintenance effort depends on what actually changed and who can judge the correct result.
AI modelsmodel customization
Assessing the business cost of custom model errors
Two models can make errors equally often and leave a team with very different work. One sends many correct documents for review. The other misses problems that emerge only when an order is being fulfilled. Before commissioning customization, a business needs to understand what those mistakes cause in its process.
AI modelsmodel customization
Training a model from scratch or adapting an existing one
A company wants its own language model, but that phrase can describe different needs: running it in a chosen environment, adapting its responses, or controlling its future development. Defining the need makes it possible to compare adapting an existing model with training one from the beginning.
AI modelsmodel customization
Does choosing a category require a text-generating model?
An application needs to assign text to an agreed category. Yet the team uses a large language model that generates an answer, which software then parses for a label. It is worth asking whether the task needs that flexibility and whether a simpler classifier could provide a useful result with less operational work.
AI modelsmodel customization
Does your Polish contact center need a custom model?
A customer tells your Polish-language contact center they no longer want their current internet plan. The representative still needs to know whether they want a different plan or want to cancel service. AI can help identify the purpose of a conversation, but a short clarifying question is more useful here than an immediate label. Consider model adaptation when clearly expressed requests keep being misunderstood.
contact centersPolish languageAI models
What a single sentiment score misses in product feedback
A customer praises the instructions and, in the same sentence, describes a frustrating changeover. A single positive or negative label tells the product team little about that experience. AI can help organize comments by what they concern, preserving both points. A custom model becomes worth considering when available tools repeatedly miss distinctions that matter for your product.
product researchB2B customer feedbackAI models
AI for catalog data: product width or package width?
A supplier lists product width and package width separately, but the import records both as “width.” A catalog specialist has to return to the spreadsheet to work out which value belongs on the product page. An AI model can help assign a supplier's description to the right field. First, check whether an import fix or an agreed mapping dictionary would resolve the error.
product dataPIMmodel customization