“How smart can we make this ghost?”
If Week 2 was about gathering context, Week 3 was about going deep. Really deep.
The design team spent the week nailing down our prototype direction and scope. This is the part of any project that feels like standing at a fork in the road with no GPS — every direction looks promising, and every choice means saying no to something else. We wrestled with big questions about what the core experience should feel like, what’s achievable in our timeline, and where the magic really lives in this concept.
We also cracked open Priestley’s original pamphlet on making carbonated water (yes, that’s a real document from 1772). Reading his actual words — his meticulous descriptions of mixing chalk with sulfuric acid, capturing gas in bladders, and shaking water vigorously — was equal parts fascinating and hilarious. The man was essentially writing the world’s first DIY soda tutorial. This primary source material turned out to be pure gold for designing an authentic interactive experience.
Programming hit a new gear with LLM Research V2. Having established that local inference was feasible in Week 1, the team now tested larger parameter models, benchmarked response quality vs. latency, and started documenting everything for the team. They also began exploring LLM customization — how do you make a general-purpose language model sound like a specific 18th-century scientist without it feeling like a bad costume? That question kept us up at night (in a good way). The possibility of starting our first real prototype was tantalizingly close.
The art team submitted their first drafts of design production, giving us our earliest visual glimpse of what this experience might look like. Even rough sketches made the concept feel more real and helped the whole team align on aesthetic direction.
On the production front, it was quarter presentation prep mode. For the uninitiated: at the ETC, “Quarters” is where you stand in front of faculty and convince them you know what you’re doing. It’s part pitch, part progress report, and part “please believe in us.” We also started building our project website — because if you’re summoning an 18th-century ghost with AI, you’d better have a good web presence to document the journey.
Week 3 was when the project stopped feeling like an idea and started feeling like a thing.
Check our quarter presentation below: