As part of the Berkley LLM Agent Hackathon 2024, this multi-agent solution was developed to solve real world problems such as: financial inclusion and financial information asymmetry, lack of computer science/programming teachers or overburdended teachers especially in developing countries such as India, help with wikipedia research, stay abreast with contemporary news and a personal Question Answering system.
This showcases the potential of making education accessible, and aid vital information dissemination.
In essence, this compound AI system showcases the benefits of creating such a system, by orchestrating several agents to solve complex tasks and accessing it all from a single screen, and all while having a conversation with it.
This system has the following features:
- Search top 5 news from hackernews and return a summary of the articles
- Search top news from the web
- Act as a personal financial planner
- Return equity, analyst recommendations, and company news for publicly listed companies in USA.
- Search Wikipedia.
- C, C++, Rust, Python Programming tutor.
- Ask me anything(AMA).
A more detailed description is here.
The system has a content moderation in place, which checks for harmful content being sent to it and asked to generate and REFUSES to answer the query.
It also has the ability to detect prompt leakage/attack and refuses to divulge any compromising information such as the API KEY, System Prompt,... .
Follow the instructions precisely.
- Navigate to
berkley_hackdir, this will be default name of the dir, if you clone this repo - Create a
.envfile and input yourOPENAI_API_KEY= - Install UV, if unavailable
- Execute
uv run src/app.py - Navigate to
http://127.0.0.1:7860 - Happy Chatting! 😉
The agent searches through hackernews for the latest articles on a topic of choice and retrieves the most recent and most relevant top 5 articles from it and summarises each article and returns the article headline, URL and the summary.
This is used for searching top news, in English, across the internet and returns top 5 news.
This acts as a financial planner and tries to guide each individual to have a financial plan in place for a better financial security and future. It can answer many personal finance questions and give recommendations based on the information you provide.
Disclaimer: All personal finance recommendations from the system are solely for demonstration purposes only. It's solely up to the user's discretion, whether to use the recommendations provided to him/her.
This feature leverages publicly available information about a publicly listed company in USA. It can retrieve the current equity price, analyst recommendations for further research, most recent company news about the said company and even compare companies tête à tête. It's an useful tool for researching about a publicly listed company before investing in it. The detailed research is also available as separate markdown files for the user to leverage.
It leverages, Wikipedia(Thank you Wikipedia 🙏), to research, simplify and summarise a particular subject, for which there's an existing Wikipedia page. If there's no Wikipedia page, then the system leverages public information from other websites and attempts to give a solution.
As the name suggests, this feature helps anyone interested in learning a programming language by providing personalised lesson plans as per the progress of the student. Currently, the tutor is confined to some of the high resource languages but it can also be used for low resource languages with further modifications to the system.
This is self-explanatory, and it aims to answer any question that the user might have, all under the ambit of the internet of course 😉.