Show HN: I transform surveys into conversation for better results

1 points by pihcys 6 hours ago

Hey HN.

A few weeks ago, I realized that the process of gathering information from my clients and colleagues could be greatly improved, leading to much better data utilization. I thought that by using LLMs, I could turn a simple survey into a conversation with an AI agent and then have the AI analyze it and present the results. After implementing this in my organization, it yielded such great results that I decided to turn it into a full-fledged product and share it with the world.

The concept of the application is simple: instead of a template database, I created a "Generate Questions" function where, after providing the context of what we want to achieve, the application creates a personalized survey that we can later edit. Upon publication, we receive a link that we can share with others to obtain responses. After the conversation is completed, the AI analyzes and categorizes its course to later assign responses to the appropriate categories. For each survey, the system creates a report that includes the results along with specific user statements to provide better context. From this level, we can access the transcript of each conversation to better analyze the entire interaction. I decided to go a step further and convert each question and answer into embeddings and save them as vectors in the database. This allowed me to create analyses where we can ask direct questions to our data to immediately get answers along with the context based on which survey responses they were generated. The application is new, and I am constantly looking for ideas and suggestions on what else I could add to further streamline this process.

Tech stack I used: Vercel AI SDK and LangChain to create AI agents OpenAI API (GPT-4o) as the LLM Postgres with pg_vector for the database along with saving embeddings.

I would appreciate any feedback.

Marcel.