Hey everyone! Quick update: I'm jumping back into coding! After focusing on leadership, it's time to reconnect with my first tech love. But here's the twist – I want to build a Python-Vue.js application with Large Language Models (LLMs), and I need your ideas.
I have a couple ideas that could be applicable, which are lower on my prioritized list of ideas to build. ;)
1) When I used to work in Accounting Software a lot, especially when working with ledger-based accounting, the vast majority of the effort seemed to be around determining the correct account allocations in order to balance the General Ledger.
I always thought it would be fun to develop a General of the Ledger, so to speak, an AI robot that could intake receipts, invoices and other financially-impacting documents and automatically allocate them in a sensible fashion.
Why it'd be cool: you could call it the General of the Ledger
How it may not apply: it's likely more of a neural-network type of thing, and less of an LLM type of thing... maybe?
2) Given the rapid adoption and use of LLMs, I think LLM tooling is greenspace. The Zed editor integrates well with ChatGPT, but leaves lots of room for improvement. Developing tooling, like templatitized prompts, and things like that, could really add value to users' lives.
Why it'd be cool: it's immediately relevant, and a lot of users could benefit from generalized tools.
How it may not apply: it likely isn't the level of LLM effort you're looking for, as it'd mainly be interfacting with LLM APIs more than anything.
3) If you were at all interested in building something that is NOT an -as-a-service, you could consider building a centralized document store using some of the offline LLMs (Llama, I think?). Surely corporations will find value in dumping all of their knowledge docs into an LLM, and being able to ask it questions about their corporate information.
You could even consider integrating with Google Docs and/or Office365 to provide the same within the cloud, but maintain corporate integrity and privacy.
Hi Tyler,
I have a couple ideas that could be applicable, which are lower on my prioritized list of ideas to build. ;)
1) When I used to work in Accounting Software a lot, especially when working with ledger-based accounting, the vast majority of the effort seemed to be around determining the correct account allocations in order to balance the General Ledger.
I always thought it would be fun to develop a General of the Ledger, so to speak, an AI robot that could intake receipts, invoices and other financially-impacting documents and automatically allocate them in a sensible fashion.
Why it'd be cool: you could call it the General of the Ledger
How it may not apply: it's likely more of a neural-network type of thing, and less of an LLM type of thing... maybe?
2) Given the rapid adoption and use of LLMs, I think LLM tooling is greenspace. The Zed editor integrates well with ChatGPT, but leaves lots of room for improvement. Developing tooling, like templatitized prompts, and things like that, could really add value to users' lives.
Why it'd be cool: it's immediately relevant, and a lot of users could benefit from generalized tools.
How it may not apply: it likely isn't the level of LLM effort you're looking for, as it'd mainly be interfacting with LLM APIs more than anything.
3) If you were at all interested in building something that is NOT an -as-a-service, you could consider building a centralized document store using some of the offline LLMs (Llama, I think?). Surely corporations will find value in dumping all of their knowledge docs into an LLM, and being able to ask it questions about their corporate information.
You could even consider integrating with Google Docs and/or Office365 to provide the same within the cloud, but maintain corporate integrity and privacy.