AI Query Generator Slack Bot for BigQuery
This app allows users to interact with a Slack bot, ask a question about the data in a table or request the table schema, and then uses the latest ChatGPT to generate a query that is executed on BigQuery to return the results. The app includes a retry mechanism for query generation in case of an error (up to two retries) and provides the LLM with the table info to generate more accurate queries. The table schema is only printed if it is successfully retrieved. All errors from retries are now passed to the LLM. The generated query is printed before the results, and the results are displayed in a pretty table format. The bot uses the Slack API to send and receive messages and parses the user's message to determine the action to take. The bot always responds in a thread to the original message.
Claude 3 Quickstart Chat API
A Chat API that allows you to immediately integrate any application you're building with the largest anthropic model (Claude Opus) with only your API key. Once you run the app, you can use this app's "/ask-claude" endpoint to send a request to Claude 3 from any application you build.
Open Source LLM based Web Chat Interface
This app will be a web interface that allows the user to send prompts to open source LLMs. It requires to enter the openrouter API key for it to work. This api key is free to get on openrouter.ai and there are a bunch of free opensource models on openrouter.ai so you can make a free chatbot. The user will be able to choose from a list of models and have a conversation with the chosen model. The conversation history will be displayed in chronological order, with the oldest message on top and the newest message below. The app will indicate who said each message in the conversation. The app will show a loader and block the send button while waiting for the model's response. The chat bar will be displayed as a sticky bar at the bottom of the page, with 10 pixels of padding below it. The input field will be 3 times wider than the default size, but it will not exceed the width of the page. The send button will be on the right side of the input field and will always fit on the page. The user will be able to press enter to send the message in addition to pressing the send button. The send button will have padding on the right side to match the left side. The message will be cleared from the input bar after pressing send. The last message will now be displayed above the sticky input block, and the conversation div will have a height of 80% to leave space for the model selection and input fields. There will be some space between the messages, and the user messages will be colored in green while the model messages will be colored in grey. The input will be blocked when waiting for the model's response, and a spinner will be displayed on the send button during this time.
Fast API endpoint for Text Classification using GPT 4
This API will classify incoming text items into categories using the GPT 4 model. If the model is unsure about the category of a text item, it will respond with an empty string. The categories are parameters that the API endpoint accepts. The GPT 4 model will classify the items on its own with a prompt like this: "Classify the following item {item} into one of these categories {categories}". There is no maximum number of categories a text item can belong to in the multiple categories classification. The API will use the llm_prompt ability to ask the LLM to classify the item and respond with the category. The API will take the LLM's response as is and will not handle situations where the model identifies multiple categories for a text item in the single category classification. If the model is unsure about the category of a text item in the multiple categories classification, it will respond with an empty string for that item. The API will use Python's concurrent.futures module to parallelize the classification of text items. The API will handle timeouts and exceptions by leaving the items unclassified. The API will parse the LLM's response for the multiple categories classification and match it to the list of categories provided in the API parameters. The API will convert the LLM's response and the categories to lowercase before matching them. The API will split the LLM's response on both ':' and ',' to remove the "Category" word from the response. The temperature of the GPT model is set to a minimal value to make the output more deterministic. The API will return all matching categories for a text item in the multiple categories classification. The API will strip any leading or trailing whitespace from the categories in the LLM's response before matching them to the list of categories provided in the API parameters. The API will accept lists as answers from the LLM. If the LLM responds with a string that's formatted like a list, the API will parse it and match it to the list of categories provided in the API parameters.
