Introduction
Your final project will be dedicated to building and deploying a fully functional LLM-powered application. Students will define a use case, integrate APIs or deploy their own model, apply prompt engineering and retrieval techniques (if needed), and implement frontends using frameworks like Streamlit or web APIs. By the end of the course, students will present a working demo and submit a brief write-up or presentation explaining their design decisions.
Goals
- Design and scope an LLM application aligned with real-world use cases.
- Integrate previously learned techniques: APIs, prompts, RAG, evaluation, and deployment.
- Optionally include tools like LangChain, Streamlit, vector databases, or evaluation frameworks.
- Deploy and present the application with a brief technical and user-facing summary.
Project Ideas
| Project Idea |
Description |
| Legal Document Assistant |
Upload contracts, retrieve relevant clauses, and chat with your LLM over legal documents. |
| Study Guide Generator |
Input lecture transcripts or PDFs and generate summaries, flashcards, and quizzes. |
| Customer Support Bot |
Multi-turn chatbot with memory and FAQ retrieval using LangChain + vector DB. |
| Scientific Paper Q&A Tool |
Ask questions about uploaded scientific papers (PDF parsing + retrieval). |
| Creative Writing Assistant |
Aid for story generation, dialogue rewriting, and tone shifting. |
| EdTech Tutor |
Personalized assistant that answers student questions and adapts tone and level. |
| Area |
Tools |
| Frontend |
Streamlit, Gradio, Flask, FastAPI |
| LLM Access |
OpenAI API, Anthropic, Cohere, Hugging Face Inference API, vLLM, Ollama |
| Chaining/Memory |
LangChain, LlamaIndex, PromptTools |
| RAG (Retrieval) |
FAISS, ChromaDB, Weaviate, Pinecone |
| Deployment |
Hugging Face Spaces, Docker, Render, Replit, Streamlit Cloud, Fly.io |
Evaluation Suggestions
- Use TruLens or Ragas for evaluating RAG workflows.
- Use prompt test cases to validate correctness, safety, and consistency.
- Track latency, prompt cost, and failure cases using logging and metrics.
Deliverables
- Working codebase (GitHub repo or hosted app).
- Documentation / README with setup instructions and example prompts.
- Optional: Deployment link.
- 5-minute video or slide demo walking through functionality and design.
Grading Criteria
| Criteria |
Points |
| Functionality & Completeness |
10 |
| Use of LLM Features |
10 |
| Innovation / Creativity |
5 |
| Evaluation & Testing |
5 |
| UI/UX or Deployment |
5 |
| Total |
35 |