Here are some random project ideas... you can use these as inspiration for your own designs.
IM in the classroom. One of the banes of professors' lives these days is people sitting in the classroom using IM to chat with each other or with other people outside the class. So, let's turn that into something useful. How would you use IM to contribute positively to the experience of a class? An IM-based tool for asking questions (perhaps less threatening to shy students than raising their hands)? An IM-based mechanism for discussions, or votes, or other aspects of classroom interaction? How could IM be used positively in a teaching or learning context, and what sorts of tools and interactions could embody these ideas?
Integrated IM. Many conversations in everyday interaction happen over or around artifacts. Meetings or conversations focus around artifacts like documents, presentations, drawings, and even sketches on the back of a cocktail napkin. In the computer, though, artifacts and conversations rarely connection. Artifacts (like files, or documents in application windows) and conversations (like IM chats) aren't related to each other; there's no way to point to artifacts in discussions, or to associate discussions with artifacts. How could we make this relationship stronger?
Persistent IM. In the everyday world, we see signs of the activities of others, persistently in the space around us. Tracks across the grass, chairs clustered together, dog-eared pages or notes in the margins of books, and even graffiti are all indications that other people have been here before us, and have left their mark.. and those marks often help us to figure out what to do and where to go. IM conversations have no persistence. How could IM be augmented with some kind of persistence? What sorts of persistence might be useful? What sorts of activities or interactions would benefit from this?
Multi-IM. IM conversations aren't like conversations in the everyday world, in a variety of ways. For one, we tent to fork "multi-threaded" conversations in which multiple different conversational topics overlap (although, as I recall, one of the Isaacs papers has something to say about this). Or again, people tend to be involved in multiple conversations simultaneously. This can often get confusing, for both participants. What kinds of solutions could be provided? (One thing to watch out for... a tool supporting multiple concurrent conversations might begin to violate Erickson's suggestion about systems that support deception. Bear this in mind!)
Specialized IM. IM tools are entirely generic. They support all sorts of people and all sorts of conversations. However, most conversations, most tasks, and more user communities have special needs. Imagine how an IM tool might be specialized to particular sorts of conversations -- conversations between baseball fans, doctor-patient interactions, conversations during meetings, etc. What sorts of facilities would be needed for specific settings? What's the best way to integrate generic and specialised IM features?