Ventura (I AM NOW AT TABLE 37)

Inspiration

I noticed how often people ask AI chatbots the same questions over and over. This not only wastes compute power and energy but also misses out on the chance to build shared knowledge. I wanted to combine the advantages of AI with the power of community.

What it does

Ventura is like Reddit mixed with ChatGPT. Users can post questions, get AI-generated answers, and continue the discussion with others or with AI. To optimize on repeated queries, Ventura suggests the top three most semantically similar posts, making answers smarter and more sustainable.

How we built it

The frontend was built with HTML, CSS, JavaScript, JSX, and React. For the backend, I created an embedding.py service powered by the FastEmbed library, which converts text into embeddings (vector representations of meaning) and compares them using cosine similarity. This powers Ventura’s semantic search, with results displayed on the frontend as percentage match scores. I originally planned to use the Gemini 2.5 API for embeddings and generation, but ran into integration issues and pivoted to my own lightweight system for a mock demo.

Challenges we ran into

Building the platform as a solo developer presented several challenges. Authentication and sign-up are still not functional through Supabase, and integrating the Gemini 2.5 API did not work in time. Despite these difficulties, I was able to deliver a working prototype and hope to gain future collaborations with it in the process.

Accomplishments that we're proud of

I am proud of turning an ambitious idea into a functioning prototype within 24 hours. Building a custom service that calculates semantic similarity was a big win, creating a testing library for filler words for optimization, and displaying those results on the frontend made Ventura more transparent and user-friendly. Most of all, I’m proud of sticking through the challenges of working solo and still delivering something that showed some progress.

What we learned

This project taught me the importance of scoping realistically at hackathons and focusing on features that deliver impact quickly. I learned how small optimizations, like trimming filler words and caching embeddings, can significantly improve performance at a large scale in the long run. I also saw the value of building in public, since even unfinished but promising ideas can attract interest and collaborators.

What's next for Ventura

Ventura is just beginning. Next steps include finishing authentication, integrating Gemini 2.5 for embeddings and generation, and improving semantic ranking for smarter results. I also plan to expand the discussion features to blend AI and community knowledge more seamlessly. Looking ahead, I hope to form a team to continue development after MHacks and eventually explore Ventura as a startup project in the future.