Exercises

Call an LLM API to summarize a wireless paper abstract in 3 bullet points. Parse the response as JSON.

Count the tokens in 5 different prompts using tiktoken. Compare zero-shot vs few-shot token usage.

Write a system prompt that makes the LLM respond only in LaTeX mathematical notation.

Implement exponential backoff retry logic for API calls using the tenacity library.

Design a few-shot prompt for classifying wireless papers into 6 categories. Evaluate accuracy on 20 test papers.

Implement chain-of-thought prompting for computing link budget calculations. Compare accuracy with direct prompting.

Build a tool-use system where the LLM can call a BER simulation function and interpret the results.

Build a simple RAG system using FAISS and sentence-transformers over 50 paper abstracts.

Compare the outputs of 3 different LLMs (GPT-4, Claude, LLaMA) on the same wireless engineering question.

Build a multi-turn conversation system that maintains context across API calls for debugging simulation code.

Implement RAG with hybrid search (combining BM25 keyword search with dense embedding search) and evaluate retrieval quality.

Build an LLM agent that can search arXiv, download papers, extract key results, and generate comparison tables.

Set up local inference with Ollama and benchmark throughput, latency, and quality against cloud APIs.

Build a complete research assistant that uses RAG, tool use, and multi-turn conversation to help design a MIMO simulation.

Implement prompt caching and measure cost savings over 1000 API calls with shared system prompts.