Answer five questions about your use case and get a concrete recommendation: whether to use prompt engineering, RAG, LoRA fine-tuning, or full pre-training — with trade-off analysis.
Question 1 of 520%
Data
Do you have proprietary domain data?
Consider: private documents, internal knowledge bases, customer-specific content, or specialized vocabulary not well-represented in public corpora.
Performance
What is your latency tolerance?
The target response time your product can tolerate. RAG and agentic flows add retrieval overhead; fine-tuned models serve faster without retrieval.
Budget
What is your operational AI budget?
Monthly API / GPU / infrastructure spend you can sustain for this use case at production scale.
Task Type
How specialized is the task output format?
Does the model need to produce general language, or highly specific structured outputs, domain-specific vocabulary, or custom behavior?
Privacy & Control
What are your data privacy requirements?
Consider regulatory requirements (HIPAA, GDPR, SOC 2), whether data can leave your VPC, and how much control you need over the model.