Maximizing Productivity: Introducing the Chrome Extension Virtual Assistant Bot with API Integration

I’m back on Medium for a reason

Calling the QnA API while handling a conversation

If you’ve been on this blog for a while or just read a few posts here and there, you might have noticed that I’m passionate about making new stuff with the code. Here’s the list of things I’ve made publicly:

• Web Scraping Script to common websites;
• Udemy course;
• Helper for Medium readers;
• SaaS that helps generate videos out of HTML/CSS scripts;
• Virtual Assistant bot capable of calling API

I’m here to present you with the last one on that list

Demo in my Twitter account

Why is it interesting?

It is based on the novel architecture developed by Meta and described in this paper. The idea is to take the most recent advancements in Language Modelling and teach them to incorporate API calls to handle conversations more naturally.

The screenshot is taken from “Toolformer: Language Models Can Teach Themselves to Use Tools”

This elegant idea allows for overcoming some of the simplest mistakes that models like PaLM or GPT3 make. For example:

• Making arithmetic calculations;
• Checking dates;
• Fact-checking;
• etc.;

The authors of the above paper thought that instead of spawning more trainable parameters in this model, they could leverage their ability to learn from a few additional prompts. But can you ask for even more?

What would you want it to do?

Even at the beginning of the evolution Regis AI solves multiple problems:

✔️ Question answering;

✔️ Helping with LeetCode tasks

✔️ Translation

Yet right now, it is in the beta version, so I’m looking for the answers to the following question:

If you’d had a bot that, in conversation, would’ve been able to order groceries or create a website (actually create, not generate script), what other things would you have wanted it to do?

If you have something, go ahead — and ask it here:

regis

1. It is free to use for the next 7 days;
2. It would not ask you for any of your data except your email;
3. For now, Regis is just answering questions to the best of his ability;

What about the engineering part?

I’ve embraced the lean startup as much as I could.

My backend is implemented using Azure Functions Apps. This is a great way for me to deploy my Python scripts without much hustle quickly.

Chrome extension is a great way of quickly presenting a completed product to customers and gathering feedback. I’ll share a tutorial on making one or two in a few days.

Last, I’m using HuggingFace inference API and OpenAI to get some intelligence behind Regis. In case you didn’t know, here’s how you can integrate any existing HF model into your app without much of a hustle. Actually, you can even fine-tune those models a bit:

Few-shot learning in practice: GPT-Neo and the 🤗 Accelerated Inference API

Summary

Well, there’s not much to say. This is a promotion post, not a tutorial you’re used to seeing here. I promise you, though, that solution behind Regis will be uncovered pretty soon.

I’d ❤️to get your feedback and determine what I can do better. Thanks for being the first adopters, and please be kind to it. 🤗