Final Project Options
COSI 112a
Fall 2025
1 Overview
The final project is an opportunity to apply concepts from this course to a substantial problem of your choosing. You may select one of the three project tracks below or propose your own project that incorporates course topics in a novel way.
2 Requirements
All projects must include:
• A 4—8 page written report (excluding references)
• Code/implementation or formal analysis (depending on track)
• A 1-page project proposal
3 Timeline
• Friday, November 14: Project proposal due
• Monday, December 2: Progress check-in
• Wednesday, December 18: Final project due
4 Project Tracks
4.1 Track 1: Research & Training
This track focuses on training or fine-tuning models to perform logical reasoning tasks using the frameworks we’ve covered in class (modal logic, epistemic logic, temporal logic, dynamic logic, etc.). Projects should involve substantial machine learning work that goes beyond parsing—your model should use these logics in interesting ways, not just convert natural language to formal syntax.
Example projects:
• Fine-tune an LLM on a modal logic task and analyze changes in model performance and behavior.
• Apply a new architecture (e.g. TRM) to SAT Solving
• Create a benchmark for evaluating temporal reasoning in language models and show how models perform on it
• Train a model to reason about multi-agent beliefs in interactive scenarios using logics we have discussed
• Compare different training objectives (SFT vs. RL) for teaching models deontic reasoning
4.2 Track 2: Systems & Applications
This track focuses on building interactive systems or applications that use logical frameworks from the course in a practical way. Your system should have a working implementation with a user interface (could be command-line, web-based, or graphical) and demonstrate a clear use case. The logic should be doing real computational work in your application—not just for decoration. Think games, tools, simulations, or interactive experiences that actually require logical reasoning under the hood.
Example projects:
• Build a multiplayer deduction game where an LLM opponent uses epistemic logic to reason about what other players know
• Build a multi-agent coalition system that computes and visualizes strategic options for boardgames or online games
• Create a dialogue system that explicitly tracks user beliefs and common knowledge given a video and prompt
• Build a deontic logic compliance checker for policies or workflows
• Create an interactive theorem prover with a game-like interface for learning modal logic
• Create a procedural puzzle generator that uses modal logic to ensure solvability and difficulty scaling
4.3 Track 3: Analysis & Theory
This track focuses on rigorous investigation of logical systems or models’ reasoning capabilities.
Projects should provide novel insights through either formal mathematical analysis (proofs, com- plexity results, expressiveness comparisons) or systematic empirical study (designing experiments, collecting data, statistical analysis). This isn’t about building something that works—it’s about deeply understanding how or why something works, what its limitations are, or how different ap- proaches compare.
Example projects:
• Conduct an empirical study of LLM performance on modal reasoning with varying complexity
• Propose and formalize a novel logical system for a specific application domain
• Apply probabilistic epistemic logic to model a real-world scenario (e.g., information diffusion)
• Investigate the computational complexity of reasoning in hybrid logics
• Formalize and prove completeness/soundness results for a novel probabilistic modal logic
• Model LLM dialogue using Kripke structures and analyze belief update dynamics
• Define a formal semantics for LLM context windows using possible worlds and prove properties about information flow
• Model multi-turn LLM conversations as dynamic epistemic logic and analyze common knowl- edge emergence
4.4 Your Own Project
You are encouraged to propose your own project that combines course topics in a creative way. You must discuss your idea with the instructor or TA before submitting your proposal.
5 Submission
Your proposal (1—2 pages, due November 14) should include:
• Title and track (or indicate if it’s a hybrid)
• Motivation: Why is this interesting? What gap does it address?
• Approach: What methods or tools will you use?
• Deliverables: What will you produce?
• Evaluation: How will you measure success?
• Related work: 2—3 relevant citations
We will provide feedback to help you scope the project appropriately.
6 Evaluation
Projects will be graded on:
• Technical quality (50%)
• Conceptual understanding (30%)
• Written communication (20%)