OneSupport

One Support

Introduction

OneSupport is a AI based re-imagination of how a support knowledge base should work, which actually helps with queries and requires minimal effort to do so. And should be simple enough to use by anyone.

Now since semantic search, embeddings, document segmentation, retrieval, llm based transcriptions, "useable" chatbots are all possible/good enough, once combined with a good looking user interface, it can be a game changer for support systems.

I started thinking about this some time back, you can see the initial ground work and thought process about this project at :

One Support

Features

  • Semantic Search

    • Using vector embeddings, and semantic search to provide better search results. This will help in providing better answers to the queries, and also help in finding the relevant documents, scripts, videos etc.

  • Chat in natural language Chatbot

    • Chatbot to help with queries. Along with a helpful answer, provides document, video, links for the reference. So user knows where the answer is coming from.

  • Scripts Script Database

    • A database of scripts, to which any one can contribute. This prevents reinventing the wheel, and helps in building a community driven knowledge base. Scripts will be used to train the model as well and thus helping in providing better answers.

  • Documents Document Database

    • A database of documents, which can be searched and used to provide answers to the queries. These can be anything pdf, txt, video etc. This will also be used to train the model, provide chat responses etc.

  • A training pipeline

    • A pipeline to train the model, and update the embeddings, and the chatbot responses, based on the new data.

  • Community driven

    • When a user submits a script or doc or a video to the database, they will have a option to keep it private, or share it with the community. If shared, the user will get a credit for the submission, and the community can upvote it, and the most upvoted ones will be used to train the model and provide answers.

  • User Interface

    • A clean and simple user interface, which is easy to use, and provides the information in a easy to understand way.

Tech Stack

  • Frontend
    • Next.js
    • Shadcn
    • Tailwind CSS
  • Backend/AI
    • Postgres
    • Langchain
    • Embedding Model(stella_en_400M_v5, jina etc, should be small yet fast & powerful)
    • LLM model(meta-llama3.18B. mistral, qwen etc)
    • Local LLM Inference(llama.cpp or vllm, LitServe, Ollama great options)