Projects
Six small projects, one per lab. Each one adds a piece, and together they become a campus helpdesk agent you can put on your GitHub and resume.
| # | Project | Session | Level | Status |
|---|---|---|---|---|
| 1 | Prompt Lab | 2. Prompt engineering | Beginner | Starter ready |
| 2 | Chatbot with memory | 3. LLM APIs | Beginner | Starter ready |
| 3 | Chat with the syllabus (RAG) | 4. RAG | Intermediate | Starter ready |
| 4 | Timetable agent + MCP server | 7. Tool calling & MCP | Intermediate | Coming soon |
| 5 | Research team (multi-agent) | 8. Agent frameworks | Intermediate | Coming soon |
| 6 | Campus helpdesk agent | 10. Mini project | Capstone | Coming soon |
Day 1: Generative AI
Project 1 · Prompt Lab
Take three everyday tasks, write a weak and a strong prompt for each, and compare the answers. Learn what actually makes a prompt good.
You will build:
- A fest caption writer (role, audience, tone, length)
- A complaint router that replies with one department name (few-shot)
- An enquiry email → JSON extractor your code can read (structured output)
- Bonus: a "CampusBuddy" system prompt, and try to break it
Stretch: test your complaint router on 20 real complaints and count how many it gets right. That's your first evaluation.
Project 2 · Chatbot with memory
Call an LLM from Python and build "CampusBuddy", a chatbot that remembers the conversation. You'll see that "memory" is just re-sending the chat history each time.
- Make your first API call and print the token usage
- Write a
cost()function and stream an answer - Keep a
historylist and send it with every request; break it, then fix it - A chat loop that shows tokens and cost after every turn
Stretch: keep only the last 6 messages, retry on rate limits, or switch the bot to Groq.
Project 3 · Chat with the syllabus (RAG)
Make the chatbot answer from college documents, with citations, instead of guessing. The starter uses a sample FAQ for a fictional college; then you swap in your own college's PDF.
- Load the FAQ → chunk it by section → embed the chunks
- Search by meaning with NumPy, then store everything in Chroma
- Answer with the source in [brackets], and refuse questions that aren't in the documents
- Measure the retrieval hit rate on a small test set
Stretch: upload your own college PDF, tune chunk size, k and the minimum score, or add memory from Project 2.
Day 2: Agentic AI
Project 4 · Timetable agent + MCP server
Give the LLM a tool. The agent decides when to call get_exam_date(subject) or a weather API, reads the result, and answers.
- Define a tool with a name, description and parameters
- Run the tool-calling loop: model asks → your code runs the tool → model answers
- Expose the same tool as an MCP server so any MCP client can use it
Stretch: add a second tool (library hours) and see the agent pick the right one.
Project 5 · Research team (multi-agent)
Two agents working together: a researcher collects facts on a topic and a writer turns them into a short report.
- Build it with an agent framework (LangGraph or CrewAI)
- Pass work from one agent to the next
- Add a human approval step before the final report
Stretch: add a third "reviewer" agent that checks the report for mistakes.
Capstone
Project 6 · Campus helpdesk agent
Put it all together. A student asks "When is my DBMS exam, and is the library open on Sunday?" and the agent answers using everything you built.
| Piece | Comes from |
|---|---|
| System prompt with role and rules | Project 1 |
| Conversation memory | Project 2 |
| Answers from the college FAQ, with sources | Project 3 |
| Timetable tool over MCP | Project 4 |
| Guardrails: "I don't know", no personal data | Session 5 |
| Deployed as an API (stretch) | Session 9 |
More project ideas for your portfolio
| Idea | Uses |
|---|---|
| Resume reviewer that scores a resume against a job description | Prompting, JSON output |
| Previous-year question paper bot: "show me all DBMS questions on normalisation" | RAG |
| Kannada / Hindi / Tamil FAQ bot for the college office | RAG, multilingual models |
| Interview practice agent that asks questions and gives feedback | System prompts, memory |
| Expense tracker that reads bill photos and fills a Google Sheet | Vision model, tools |
| GitHub repo explainer: "how does this project work?" | RAG over code, agents |