About Kumori

Kumori (ku-mo-ri / 曇り) means 'cloudy' in Japanese. So this is a bit of play on the word where instead of grey and gloom of clouds (and this being cloud-based), it's been said to embody the sunshine and positivity that comes after the clouds and rain --and our goal is to provide clean/concise data in a similar fashion via cloud!

A bit about us...

Robots can make content, but lack your style. Busy professionals don't have time to keep up with every new storylines that comes across. However, it'd still be valuable for them to comment/post about it to their audience. Kumori can help. We'll watch news stories based on your preferences, give you a dashboard of the likely top storylines you care about, then get you started with content using your voice, based on the sample writings you've given us.

An enormous (~106M tokens across multiple models) amount of OpenAI's new Assistant stack, mixed with some Claude and Google infrastructure to power the backend.

A bit about the tech:

Flask, a micro web framework written in Python, serves as the backbone, and Bootstrap for the UI.

The backend stack of Kumori is everything Google. It leverages Google Cloud's versatile infrastructure to provide a fully integrated backend experience: Cloud Storage PostgreSQL serves as a reliable and scalable database, while Cloud Tasks efficiently manage background processes. We utilize Google Cloud Storage for robust and secure object storage, ensuring your data is always accessible. Additionally, Cloud Functions and Cloud Scheduler automate nightly jobs to maintain system integrity, and Google's IAM safeguards credential management, keeping your interactions with Kumori both seamless and secure.

Auth0 streamlines user authentication for Kumori, offering a smooth and secure login experience with multiple methods, ensuring that access to personalized content creation tools is both elegant and robust. This integration underpins our commitment to protecting your identity and preferences with ease of use in mind.

Kumori utilizes gpt-3.5-turbo for free users. It provides quick and efficient initial content prompts while providing enhanced, detailed drafts through GPT4+ for logged-in users, ensuring clarity and consistent uptime in our AI-powered content generation platform. This tiered approach optimizes both resource use and user experience, tailoring the AI's depth to your engagement level.

Dalle3+, currently the main ingest integrated into Kumori. However, have some cycles into harnesses advanced image generation models like Stable Diffusion and RealVizXL from HuggingFace, to further enhance your posts with rich, AI-crafted visuals that resonate with your brand's narrative. Our stack is optimized to deliver uniquely customized imagery that elevates your storytelling on-the-fly.

Mediastack sources from 7500+ news sources and enables Kumori to seamlessly aggregate tailored news content relevant to your interests and style, providing a foundation for AI-driven drafts and image suggestions that match your digital persona. Our users get a curated feed of real-time news, which our AI refines into engaging posts and complementary visuals, saving time while preserving uniqueness.

Socket.io empowers Kumori with real-time interactive capabilities, allowing for instant delivery of ChatGPT responses, ensuring a seamless conversational experience in your content creation process. It facilitates rapid communication between server and browser, so your AI-generated content is as dynamic and responsive as you need it to be.

A bit about what's next:

v1.0 rollout 2024March22. This is the first iteration and there's been good feedback for new additions. Take a look at the 'Tomorrow' tab below for a few of them. If you have an idea for something new, linkedin.com/company/kumori-ai/

There's been a few things pitched, and the list grows, but here are some potentials:

  • Machine Learning patterns to accept multiple samples to distill closer person.
  • Creating the storyline and image automatically to have a dashboard just to review (cost-prohibitive for v1 though).
  • StableDiffusion, SDXL, RealiViz via Huggingface/Diffuser for varied image generation.
  • Utilization of expanded LLMs for more comprehensive language capabilities and to give users a choice which to use.
  • Alignment to other social media platforms for more laser focused audience engagement.
  • Fine-tuning of the models used for each API call based on user preferences.
  • Addition of new media outlets to the news monitoring repertoire.

Assembling silver linings in clouds, un momento...