ai projects
Our endeavors have led us to architect innovative AI tools that promise to elevate our capabilities across various domains – from content creation and data management to art and the very genesis of ideas. Imagine chatbots morphed by the brushstrokes of artistic intent, or voice assistants tuned to resonate with the depth of creative thought – these are not mere tools but collaborators, designed to weave through the layers of complexity and simplicity in an elegant dance of innovation.
This page contains a summary of our recent efforts to create these AI tools.
asyra ai chatbot
Our AI chatbot, Asyra has been a staple in our creative process for a while now. While there have been articles discussing model training techniques and how we use more unstructured data to provide context, we are yet to breakdown how exactly we created Asyra.
A Vector Database of Articles, Content and Ideas
As discussed in the article, AI Chatbot Training: Your Own Thoughts - Asycd, we try to add various ideas from myself and other sources to the vector database of context to be injected into prompts. Aside from ideas and thoughts of the founder, we have all the articles, documentation, plans and more added to the vector database which allows us to have extensive context about Asycd. This comprehensive context empowers us to generate prompts that are not only relevant and insightful but also aligned with the unique identity of Asycd.
Similarity Search + Keyword Extraction
It was essential that we obtained the relevant context from the vector database and our method of doing that involves keyword extraction. Using PyTextRank, which is a an algorithm Google uses rank text in terms of relevance, we was able to make the similarity search more effective. Below is a snippet of the function used to perform the similarity search.
It begins by extracting keywords from a given query, which can be thought of as a distilled representation of the query’s main points. Then, for each keyword, it performs a separate similarity search and then results for each keyword are aggregated into one string. This combined with the results from the entire query similarity search.
By combining the retrieved documents related to both keywords and the query itself, a richer context is formed. This context is then injected into a prompt to be used to help respond to my query.
Other Features of Asyra AI
Context-Aware Responses: By retrieving context from a vector database, Asyra provides responses that are relevant to the current conversation thread and user history.
Real-Time Communication: Asyra utilizes OpenAI's GPT model to engage in real-time dialogue, ensuring prompt and dynamic interactions. Asyra also has access to the time and can make measured judgements on data based on the time difference.
Brand-Aligned Communication: Trained on Asycd's brand data, Asyra maintains a consistent voice that resonates with the brand's values and style guides.
User Engagement Analysis: Asyra learns from user engagement metrics to continually refine content recommendations and interaction quality.
Note: These features are as described by Asyra AI and may slightly differ from real-world capabilities
ai voice assistant
At Asycd, we're pioneering a unique, gender-neutral AI voice assistant, “Asyra”, crafted from the ground-up using my own voice to seamlessly assist you with daily tasks, content creation, and the exploration of new ideas. While Asyra's potential applications span a multitude of domains—from personal education to financial guidance—our current commitment is to solely enhance the Asycd experience.
Similar to the chatbot version, it has access to a vector database of our documentation, thoughts and ideas, as well as data to help provide contextual responses in an auditory fashion as opposed to just text. Below are some sample responses by Asyra: