search

JedWalton / lucidify

File Q&A with Vector Databases along with a ChatUI and bespoke datasets. Golang, Nodejs, Python, Weaviate, OpenAI, Postgresql, Automated testing, Docker.
1 stars 0 forks source link

Product ideas #19

Open JedWalton opened 11 months ago

JedWalton commented 11 months ago

Idea iteration 1:

Product is a real time support assistant to make handling Customer Success Manager for SaaS products that much smoother.

ChatGPT Threads Appear with Real conversational threads. Our AI will Search our corpus for relevant solutions. Preload Ideal response. Customer Success Manager can then manually approve suggested response.

We have a Vector Database. We have a Chatbot UI. Product uses openai to manage threads.

We can manipulate threads with our own inputs from various data sources.

We can create buttons to take various actions.

Essentially Human Supervised/Human-in-the-loop Autonomous support agents.

Some useful resources:

https://www.lennysnewsletter.com/p/hiring-your-early-team-b2b https://www.reddit.com/r/SaaS/comments/174scom/who_are_you_saas_founders_hiring_first/ https://www.reddit.com/r/startups/comments/1122l74/what_should_a_startups_first_hires_look_like/

Anyone dealing with large numbers of support tickets.

This UI could be a niche tool to help them serve higher volumes of support tickets per Customer Success Manager.

JedWalton commented 11 months ago

Vector Database Class can be created with existing support tickets, and solutions. If the vector similarity is above X, then respond with high confidence the proposed solution is accurate. If lower confidence, then require manual support tickets.

Correctly identified support tickets increase in weight over time. The language model is the bridge between customer and vector data store.

JedWalton commented 11 months ago

Idea 2: Curate corpus of data for all companies on linkedin. Use NLP to categorise all into niches. (More granular the better) Scrape all websites docs etc and vectorize to be able to vector search knowledge per company.

Scrape profiles and save email addresses per niche. https://github.com/TufayelLUS/LinkedIn-Scraper

Use vectorised data per company to generate cold outreach via openai to each potential client.

Close sale to scheduled warmed lead to calendar.

Sell warmed leads.