I am looking for a developer who can help me to plug a number of ai tools together to create a new service. I am new to this world but hoping someone might be able to help me or point me in the right direction of those who can !
many thanks.
1 Like
Hey champ!
And welcome to the developer community forum. Just tell us some more about what you want to create, and I’m sure people will point you in the right direction. 
2 Likes
Some idea of renumeration would also be welcomed, as exciting as your venture might turn out to be, equity alone is often frowned upon.
So in short I would like to create a tool that enables me to undertake 1-1 interviews with real people at a mass scale (000’s) at the same time and then analyse those conversations to identify the key trends and themes in them. This means we would need to be able to design a conversation structure, conduct the conversation with the participants (sourced from paid for lists), capture the conversations and then analyse them. In terms of renumeration, I’m super happy to offer equity and a day rate to cover immediate time as well as a longer term pay out should we succeed. (which we will!)
3 Likes
As a general note, finding trends in large datasets is still a challenge with limited context AI’s, this kind of data processing requires more analytical methods to be employed. It is firmly in the realm of data science.
AI for unstructured to structured conversion is the main use case presently, you could certainly use some of the features associated with current gen AI’s such as embeddings to find clusters of common semantics, but again that is only a small part of finding trends in human generated data.
I’d estimate 10,000 - 15,000 hours as an entry point, you’d need data scientists and AI specialists, and a project manager. Depending on how the package would be used, you’d also need UX/UI designers, networking specialist, front and back end programmers and possibly a database specialist, it’s a sizable project.
2 Likes
Hi Ben,
I have some ideas on where to start. Most of them I’ve already published here. If you need more detailed conversation, I’m open for consulting.
Basically the approach I took was too break down the whole process of reading/comprehension humans use, and then implement them as separate tasks chained and combined with regular programming to structure the otherwise raw text string.
As for the currently mentioned use, after the text structuring you would need to pass the resulting object to a set of user intent/conversation workflow analysis tasks to get the “pattern” of the conversation itself.
I think the most difficult part of the latter is that none really knows what to analyse in conversations to get a general "pattern"applicable to all conversations and usable as base for further detalization… Maybe some philosophers would help here more than engineers.
1 Like
The structure or pattern, of the conversation is already determined by the interview format:
But I 100% agree with @Foxabilo, this is a sizable project, not a 1 person gig.
2 Likes
This is the raw fact, Totally agreeđź‘Ť
1 Like
Well, the basic structure is not a problem here (currently in that part I can cut to specific subjects inside a single speech block with parent/child relationships between subjects), that one is rather easy.
The patterns I was talking about are more like “interview script template parts” where the machine analyses speakers intent and conversation road map. Something like:
Speaker 1 tried to sell a product for x, but potential buyer hesitates about the result but prefers to mention price concerns only…
I think that’s the true goal of the thread? Isn’t it?
A way for a machine to read what people want and how they react in a conversation with all the possibilities it opens for automations.
Thanks for all of your thoughts. I agree it’s not a 1 person job. And yes, the structure of the interview would be pre-defined - to cover a set of questions. that you can then analyse based on those predetermined parameters. The big unlock however would be for the platform to be able to react to the respondent and tailor the conversation accordingly - so it becomes very human in it’s ability to react and respond to where the conversation goes. I would also like this to be audio not text based - so it’s like having a phone conversation, not engage with a text chat.
1 Like
I think the audio part of it, while not quite trivial, is a great deal simpler to implement (latency notwithstanding)
Bringing dynamic responses into the mix brings this outside of what is currently possible with this generation of LLM’s. Next generation with significantly larger contexts and running on more powerful hardware would probably make this a reality. At that point this becomes a curated conversation between an AI with a task and goal and a human, with a central AI overseeing all of the conversations and creating maps of interest, then you can query the AI on the current trends from all of the current conversations.
2 Likes
That part is sort of “easy” as more or less technical.
The most difficult part here would be there real-time conversation leaded by a machine to get to the predefined goal while adapting to human responses and reactions…
How many records of such conversations you have to see what’s possible?
Have you thought about how you usually break down the conversation/interview in this specific domain? (Might be worth spending some time with top-calls guys in this domain and see how they draw the conversation outline)