Techniques
List the LLM techniques you intend to cover in your set of use cases. A typical list would look like: 1) Prompt engineering, 2) LLM compression techniques, 3) Unimodality, 4) Multimodality, 5) Data analysis, 6) Synthetic data generation, 7) Multilingual capabilities, 8) Multi-agent frameworks, 9) Code generation, including User Interface creation and application building, 10) MCP servers, 11) Under-represented languages, 12) Guardrails, 13) Benchmarking of the LLM and benchmarking of the LLM-based application you have built, 14) Retrieval augment generation (RAG), 15) RAG output evaluation, 16) Agentic AI, 17) Token optimisation techniques, 18) Inference engineering, 19) Supervised fine-tuning (SFT), 20) Parameter efficient fine-tuning (PEFT), and 21) PDF document handling and embeddings.
What to build?
List solutions you can build to solve business problems: 1) Chat with website, 2) Customer-facing chatbot, 3) Quiz, 4) Survey, 5) Translator, 6) Question-answering system, 7) Learning through story generation, 8) Website, 9) Automate test cases, 10) Chat with content repositories. Honestly, this list will be limitless. I have mentioned just a few examples to guide your thought process.
Which industry are we solving the problem in?
The list would most likely look like: 1) Financial services, 2) Healthcare, 3) Manufacturing, 4) IOT, 5) Banking, 6) Insurance, 7) Retail and e-commerce, 8) Education, 9) Transportation and logistics, 10) Media and entertainment, 11) Gaming, 12) Government, 13) Agriculture, 14) Legal, 15) Oil and gas, 16) Real estate, 17) Hospitality, 18) Automotive, 19) Pharma and Biotech, 20) Construction, 21) Fashion and apparel, 22) Fintech, 23) Cybersecurity, 24) Genomics, 25) Utilities, 26) Clinical psychology, 27) Airlines, 28) Chemicals, 29) Human diseases, 30) Telecommunications, 31) Cryptography, 32) Weather and climate, 33) Space mining, 34) Journalism, 35) Embedded software, 36) Defense, 37) Mining, 38) Maritime, 39) Consumer goods, 40) Food and beverage, 41) Media publishing, 42) Sports, 43) Travel and tourism, 44) Nonprofit, 45) Professional services, 46) Accounting and audit, and 47) Renewable energy.
Which function are we solving the problem in?
The list would be: 1) Marketing, 2) Sales, 3) Human resources, 4) Supply chain, 5) IT, 6) Finance, 7) Service assurance, 8) Ordering and fulfilment, 9) Billing and payment, 10) Operations, and 11) Research and development.
Framing a use case
Pick one technique set: PDF document handling, embeddings, and RAG. Pick one from what to build: Question-answering system. Pick one from industry: Agriculture. Pick one from function: Sales.Now you have a use case: Build a question-answering system that responds to user queries related to the sales of agricultural products using an LLM, PDF content embeddings, and RAG. Note that it is not possible to employ only one Generative AI technique for one use case. Multiple techniques work in unison for a single use case. For the above example, one would need prompt engineering, embeddings, RAG, multimodality (if using image, voice, or video capabilities), inference engineering, token optimisation, optionally a compressed LLM, guardrails, and solution benchmarking, all at once. However, your use case can focus only on one aspect of the technology in one use, for example, only prompt manipulation or prompt study.Therefore, how many use cases can we frame using just the above information? Theoretically, it will be 21 × 10 × 47 × 11 = 1,08,570 (One lakh eight thousand five hundred seventy). Pick any one of them and work your way through building an LLM-based solution.This combinatorial framing demonstrates that Generative AI is not limited by ideas, but by execution discipline. With over 100,000 potential use cases, the real leverage lies in selecting one well-scoped problem and rigorously engineering it end-to-end. Start small, iterate with measurable benchmarks, and expand systematically, because scalable impact emerges from depth, not breadth, in applied AI systems.
