How to create AI Chatbot

With ChatGPT coming to the stage, we entered a new era of Generative AI development. OpenAI API brought significantly updated ways of AI text generation integrations and learning of LLMs.

Springs, as a one of the software AI development companies, stayed within it, and we became one of the pioneers in the new way of AI chatbot OpenAI development. Let me share our story about ioni.

Choosing a Vector Database

We started our journey with a simple POC, trying to get the best answers from our human-like chatbot, and one of the difficulties we initially met was choosing a vector database.

Vector DB is a specialized form of an AI database designed to store information in the form of high-dimensional vectors. These vectors serve as mathematical representations of various features or attributes. Milvus, Pinecone, Weaviate, Singlestore — these are really popular vector databases. But we decided to go with Chroma — the AI-native open-source embedding database. Why?

ioni is fully based on LLMs (large language models) development and embeddings, so Chroma was the best fit for us from this point. Chroma makes it easy to build LLM apps by making knowledge, facts, and skills pluggable for LLMs.

Used LLMs

A large language model, LLM is an advanced machine learning model capable of executing a broad range of natural language processing (NLP) tasks. These tasks encompass generating and categorizing text, engaging in conversational question answering, and facilitating text translation between different languages.

ioni.ai uses state-of-the-art LLMs developed by top providers such as HuggingFace (open-source), OpenAI models, and Cohere. In technical terms, we employ embeddings for AI-powered search capabilities and prompt engineering for effective text manipulation.

An LLM Chain is a chain that adds some functionality around LLMs. It is used widely throughout Python LangChain, including in other chains and agents. It consists of a Prompt Template and a language model (either an LLM or chat model). It formats the prompt template using the input key values provided (and also memory key values, if available), passes the formatted string to LLM, and returns the LLM output.

ChatGPT Integration

As far as ioni is based on ChatGPT4 API integration, the OpenAI solution played a key part in our AI application.

OpenAI’s generative pre-trained transformer model has undergone extensive training to grasp both natural language and code. These models are capable of generating text as output based on the given inputs. In the context of GPTs, the inputs are commonly known as “prompts.” Developing a prompt essentially serves as the way to “program” a GPT model, typically involving instructions or illustrative examples to guide the model in successfully accomplishing a given task.

Recently, OpenAI released the newest version of ChatGPT4, which allowed us to define functions for gpt-4–0613 and gpt-3.5-turbo-0613, enabling the models to intelligently generate a JSON object comprising the necessary arguments to invoke those functions.

These models have acquired the capability to identify instances where a function invocation is required based on user input. They can then generate JSON responses that align with the function signature. By incorporating function calling, we gained a more dependable means of retrieving structured data from the model.

MVP

Our Minimum Viable Product (MVP) is a web application created using Python and Javascript. It incorporates a built-in AI algorithm based on ChatGPT4. The integration of ChatGPT API enhances the application’s capabilities, enabling seamless and efficient data processing and retrieval.

The primary product feature revolves around empowering users to generate an OpenAI Chatbot Widget and embed it onto company website using a straightforward code copy-paste option. This functionality grants users the convenience of implementing an AI-powered chatbot on their website with minimal effort and technical complexity.

MVP version also allows users to add data to the knowledge base and improve the quality of answers in such a way. You may use the section “Add data” to add any HTML or text pages to the AI database.

What’s Next?

OpenAI, along with other providers of large language models (LLMs), continues to showcase remarkable advancements in machine learning. Given the rapid progress in this field, it is highly likely that we can anticipate the emergence of new AI algorithm features in the near future.

Springs team consistently monitors new updates and incorporates the latest AI features into our products, enabling us to enhance our OpenAI Chatbot GPT builder with cutting-edge technologies on a regular basis.