Final Project

The final project is the centrepiece of this module (100% of grade). You will experiment and build a complete intelligent system demonstrating your ability to address a narrow research question and design and implement a foundational model-based application.

Project Topics

A list of projects will be shared on Canvas. You may select one of the provided topics and outline your strategy in your proposal. If you prefer to pursue your own idea, you must obtain the instructor's permission beforehand via. email.

Requirements

Projects must be completed individually. Each project must address 2-3 interesting narrow research questions by conducting comprehensive experiments. For instance, a narrow research question could be designing an evaluation for an aspect of your task and exploring the impact of model size and context window length on performance. In addition, your project must utilize all of the following:

• Prompt Engineering OR Appropriate Context Construction
• RAG OR Agentic Capabilities (tool use, planning, etc.) OR Multimodal (vision, audio)
• Fine-tuning (parameter-efficient instruction tuning is required; the use of RLHF will receive bonus credit)
• Local Inference
• A Small Front End for Demo

AI Tools Usage Policy

Please refer to the AI Usage Policy and Academic Integrity section on the policies page.

Timeline

Available Resources

Students will have access to the following resources for their projects:

• Lab PCs: Alienware 16X Aurora AC16251 — NVIDIA GeForce RTX 5060 (8 GB GDDR7), Intel Core Ultra 9 275HX (24 cores, up to 5.4 GHz), 32 GB DDR5 RAM, 1 TB SSD. Available all day on Wednesdays; access on other days can be arranged on request.
• AMD Compute & AI PCs: Access to AMD compute infrastructure and AI PCs for projects requiring larger-scale GPU resources
• API Access: API access to a wide range of large open-source foundation models, thanks to our sponsors