Description:

Modern applications are expected to understand humans and also respond like one. Natural Language Processing (NLP) is one of the key technologies in building such features. Some of the most common NLP tasks are sentiment analysis, named entity recognition (NER), language translation, and text summarization. These tasks have been around for decades but traditionally required a data scientist or machine learning expert to curate and label data and train a deep neural network which can easily take a couple of months.

With the advent of Large Language Models (LLMs) like GPT, these have become accessible to developers and can be built in days or a couple of weeks using Prompt Engineering and Retrieval Augmented Generation. While it has become a lot simpler, it still requires care and caution for response to be reliable and effective.

In this talk, you will learn to build an end-to-end multi-lingual virtual assistant that takes voice input and responds intelligently.

Outline:

• NLP Tasks with GPT (15 min)
  • GPT Setup
  • Sentiment Analysis
  • Language Translation
  • Named Entity Recognition
  • Text Summarization
• Backend Microservices (7 min)
  • Use FastAPI to stitch all NLP tasks into REST endpoints
• Frontend (3 min)
  • A simple browser-based voice-enabled chat interface
• Q&A (5 min)

Takeaways:

• Gentle intro to common NLP tasks
• How to write effective prompts for extracting structured info from natural language text
• Structuring LLM-based features into microservices
• Joy of building end-to-end intelligent LLM-based virtual assistants from scratch

Prerequisites:

Basic knowledge of Python, asyncio, and microservices.

Speaker Info:

Satish Chandra Gupta is a seasoned Data/ML Practitioner and Advisor/Consultant with over 25 years of industry experience. He was co-founder and Chief Data Officer at Slang Labs, where he helped create a Conversational AI platform for building multilingual, multimodal voice assistants in mobile and web apps. Before that, he was building real-time streaming data pipelines and ML applications at Ola processing a billion events a day. Prior to that, he worked at top-tier organizations such as Amazon, Microsoft Research, and IBM Software Lab.

His expertise lies in helping startups and small to medium-sized companies craft a data/ML strategy and build an economical low-maintenance data infrastructure, actionable data analytics, machine learning, and LLM applications. He shares his learnings in a popular newsletter Machine Learning for Developers (ML4Devs) with over 8,000 subscribers, and on LinkedIn with over 26,000 followers.

Before his foray into data and ML, he built compilers, profilers, IDEs, program analysis, and developer tools. He earned a BTech in Chemical Engineering from IIT Kanpur in 1996 and an MS in Computer Science from the University of Wisconsin - Milwaukee in 2001.

Speaker Links:

• Blog/Newsletter: ML4Devs - Machine Learning for Developers
• LinkedIn: scgupta
• I regularly speak in Meetups and Developer events in Bangalore. Here are links to selected previous talks:
  • PyCon India 2019 Workshop: Building, testing, and profiling efficient micro-services using Tornado
  • TFUG BLR, Feb 2020: Everything Speech! - Machine learning applications in Voice Technologies
  • MLOps Community Podcast, Oct 2020: Data Engineering + ML + Software Engineering
  • Google DevFest BLR 2022: Tradeoffs in building Big Data Pipeline on Google Cloud (Day 2, Track 1, Video)
  • TFUG BLR, Oct 2022: BigQuery ML: Data Processing and Machine Learning at Cloud Scale
  • TFUG BLR, Dec 2022: Vertex AI for Beginners
  • TFUG BLR, Sep 2023: Keras Community Day: Emerging Architecture and Design Patterns for LLM Applications