Building with GenAI #2

Challenges, Learnings & a Bit of News from the Builders of Quivr.app

Building with GenAI

Challenges, Learnings & a Bit of News from the Builders of Quivr.app

New features and learnings from the team at Quivr and news from across the exciting world of Generative AI.

Quivr is sponsored by Theodo, Aleios, Padok, Sicara and BAM. Resources are shared from these companies to work on the project. Outside contributions are always welcome and we have a growing network of active contributors.

Last week at Quivr

  • Last week in Numbers

    • 6 Releases: v0.0.30 to v0.0.35

    • 10 Active Contributors this week

    • 1800 new users

    • 8.5k Questions asked

    • 14k Files Uploaded

    • 6K new Github Stars

  • New Features and Improvements

    • Search through your brains

    • Multithreading for database calls

    • Increased frontend tests

    • Backend refactored to facilitate integration from new LLMs

    • Static analysis of backend types

    • Authentification with Google

    • Use of security headers

    • A bit of prompt engineering

Demo: Quivr.app - Source Code: GitHub

Learnings from the week

  • Stan - Sensei: The value of a good product will come from the underlying prompt. Prompt engineering is important! Analysis tools such as Smith from Langchain are critical 🛠️

  • Zineb - Chief Dev: As Quivr grows and evolves, so do our priorities. Continually balancing security, privacy, feature delivery, stability, and the complexity of AI-generated qualitative answers is top of the mind 🤯

  • Mamadou - Frontend Wizard: From TS, static typing gives me more confidence 👍 Tests should be simple and ideally follow the Single Responsibility Principle (SPR). When the test starts to have too much overhead, it's time to review your architecture 🧪

  • Matt - Backend Dude: People are asking for a solution to keep data private and on their own system. This will be a high priority moving forwards. All these tools are still very young!! I broke prod by upgrading to a newer version of Langchain ☹️ Rookie error 🧐

  • Chloe - Product Maestro: Great UX is crucial! It ensures that the added features meet the users' needs in an intuitive way. We can develop relevant and user-friendly features that provide a smooth and enjoyable experience 🤣

  • Brian - Methodology King: Acquisition is hard, retention is harder. Less than 2% of users have Quivr in their daily habits... Deep diving in the analytics helps cracking the user engagement along with product improvement 🚀

Guest Entry: Chantal - UX auditor from BAM
  • I liked the simple and clean look. It's in line with what's trending for these kinds of things, and it puts you at ease.

  • When I scrolled down to the bottom of the homepage, it’s kinda bare, and it felt like they were only talking to developers (the fact you can only reach them on Github and Discord).

  • I ran into a few bugs and weird stuff happening.

  • To be honest, I wasn't really sure what to do next (they use a lot of jargon and don't really give you any tips on what to do or how to use it.)

Thanks Chantel for your feedback 🤩

News from the World of Gen AI

  • 🎉 Open AI’s GPT-4 is now generally available for all users of the API! See the press release here

  • 🚀 More big news from Open AI with their new Code Interpreter now being rolled out to all ChatGPT Plus subscriptions.

  • 💫 Anthropic have just announced their new model Claude 2. This aims to compete with Open AI’s chat and API offerings. Up to 100k tokens can be used for context, trained on data up until early 2023.

  • 🧪 Tabnine just released Tabnine Chat in beta. It allows you to interact with Tabnine’s AI models in a flexible free-form way, using natural language from inside the IDE.

  • 🏆 Last but not least, Quivr has been on the top of the Github trending page for 2 days now!! Thanks everybody for your support and help 😍 We are just getting started 👀

Jump over to Quivr to chat with your second brain for free 🧐 Head to GitHub for the code and give the repo a star ⭐️ while you’re there 🙏