Kindred

• 

  Kindred

Inspiration

Information overload is a real problem today, many people want to stay up-to-date with the latest tech news but can't.

Modern LLMs have made it possible to summarize information at scale with a very high level of quality.

What it does

Kindred summaries HackerNews posts and their related information. These summaries are then displayed on a web page, make it fast and easy to stay up-to-date with the latest HN posts.

Each post summary includes links to both the story's URL if present and the original post on HN. A list of key insights is presented per post. Clicking an insight shows its most relevant comments in-line, each clickable to view the original on HN.

You can also chat with Kindred about a post summary to ask any questions you may have about the story, the post and its comments.

Easily specify your interests and tastes, which can be related to or independent of the social media site summarized. The system uses this information to rerank the post summaries according to each user's interests.

Tastes include books, movies, podcasts and more. Recommendedations for tastes are also sourced once any tastes are initially added, via Qloo's insights API.

How we built it

The tech stack consists of:

• Next.js + TypeScript
• Apollo GraphQL and Prisma (Node libraries)
• Serene Core (Node/TS basics library I wrote myself)
• Serene AI (Node/TS AI library I wrote myself)
• PostgreSQL (+pgvector)

Integrations:

• Algolia HN API
• Qloo (Tastes AI platform)
• LLMs (configurable)

Challenges we ran into

I was new to Qloo and its API, so this was a learning experience. But the docs are quite good, so it wasn't difficult.

Accomplishments that we're proud of

Kindred uses Qloo in a way that makes excellent use of its tastes data.

The system ties LLMs and Qloo together in a way that makes reading social media posts in summary form, ranked by user interests. It's much easier and faster to stay up-to-date this way.

What we learned

I learnt about Qloo, its concepts and how to use the API to get relevant data.

What's next for Kindred

I want to continue to host and improve Kindred. Although limited to HackerNews only, it can be expanded to summarize other social media sites like Reddit.

Built With

• algolia
• apollo-graphql
• llms
• next.js
• pgvector
• postgresql
• prisma
• qloo
• typescript