AI Notes - AlphaSense and Future Home Grown AI Tools

In the past month, I've written a few investment memos from soup to nuts, so now have strong opinions on what LLMs might be good for. 

First, AlphaSense. We have access to AlphaSense, so I've been able to give it a whirl. My takes:

• Mediocre but voluminous investment memo writer -- We've come up with a decent investment memo prompt, and it can pull together a 20-page investment memo in ~15 minutes. However, some info isn't relevant, some isn't well supported, and overall ... it's 20 dense pages. More detailed gripes below.
• References to internal sources -- One thing I love and hate is that it links to internal materials (e.g. Affinity, Sharepoint). Upside: when I've had trouble finding a single piece of info from the data room, AlphaSense did a good job on hunting down the sources of truth. Downside: it'll link to old memos or in-flight memos, which is not helpful if I'm trying to verify or double-check details for the in-flight memo. 
• Expert calls -- I've been diving into some topics I have little prior knowledge in, so I've tried to make use of what I see as AlphaSense's primary edge: expert calls. I've tried to ask it something like: "Find me all expert calls that talk about X industry or Y company or its competitors" to get better situated from insiders. Overall valuable, but not worth the price of AlphaSense alone.

My current thoughts on tools that'd be (a) buildable and (b) useful:

• Draft investment memo writer -- I think a well-constructed LLM tool can write a good investment memo. The keys are (a) asking the right sequence of questions to the LLM, (b) linking out to external sources for some pieces (e.g. for public market comps), and (c) knowing which pieces to delegate to LLM and which pieces should be 95% human. (I'll dive more into in another post, but one example: the "Investment Thesis" section should 100% be human-written -- if not, what is your value as an investor??)
• Data room analyzer -- I wrote a little tool to help spill out all the contents of the data room (i.e. show me all the files in all the folders in an easy way), which is a good first start. Next step would be to get a rough summary of each files, and step after would be doing some heavier analysis. It'd also be great to be able to ask questions of the files, or at least have it help me find where X company is referenced. (i.e. what I use AlphaSense for). Should be buildable with LLMs and RAG (at least in a rudimentary form) -- big question is if 80% of max capability (what I-plus-AI am capable of) is good enough.
• Investment memo cleaner upper -- Would love to have a tool to go through a memo and tidy up tables, help with resizing, flag areas to review/condense, etc. Not sure how much you can integrate into a Word doc. Will explore.