Concept map: Prompting vs retrieval vs fine-tuning

Taught and used in

LLM architecture and adaptation, Business judgment

Insights 19

• Supply volatile facts at request time because parametric knowledge is frozen at training
• For fixed-label text classification, benchmark a small encoder before a generative LLM
• Match checkpoint type and chat template to the job before judging an open model
• Use few-shot examples to show format and label space, not to teach new rules
• Supply context instead of personas, because role prompts do not add knowledge
• Require a fine-tune to beat the best prompt on a frozen eval set before you own it
• Name the missing capability before choosing prompting, retrieval or fine-tuning
• Do not fine-tune to inject facts that change; retrieve them instead

Projects 7

• Zero-shot time-series foundation models vs strong baselines on post-release monthly data (docs-only)
• Flagship: passenger-rights document Q&A with a retrieval baseline, one controlled improvement and an honest generation mode
• LLM-as-judge reliability study: agreement with human pairwise labels, position bias and length bias
• Golden-set regression suite that gates prompt changes for a claims-intake triage assistant
• Summary faithfulness evaluation for legislative bills: human claim labels vs automatic checks
• LoRA/QLoRA adaptation plan for fine-grained banking intents (implementation-ready, docs-only, GPU optional)