Connect GenAI models to the outside world (By: Ehtisham Raza) - DevFest Lahore 2024

Talk by Ehtisham Raza (https://www.linkedin.com/in/ehtisham-raza-0a5479163/) at DevFest Lahore 2024 by GDG Lahore.

Transcript

1. Connect GenAI Models to the Outside World Lahore Ehtisham Raza
2. Ehtisham Raza AI/ML Engineer and Data Trainer & Coach at Data Techcon I am AI/ML Engineer working in a US Based Startup having 5+ years of experience and solving financial and litigation problems using artificial intelligence and machine learning.
3. Introduction to Generative AI Importance of External Connectivity Overview of Google Gemini Integration Methods Google Cloud Tools for Integration Agenda
4. What is Generative AI? Unstructured data Output Input GenAI Model New Content • Generative Al refers to a set of artificial intelligence methodologies that can produce novel content that resembles the training data they were exposed to. • The content could be anything spanning from synthesizing text, generating code, realistic images, music and more
5. What is Generative AI? Output Input GenAI Model New Content
6. Overview of Google Gemini • Successor to PaLM 2, developed by Google DeepMind. • Multimodal capabilities: processes text, images, audio, and video. • Variants: Gemini 1.5 Flash, Gemini 1.5 Pro, each optimized for different tasks.
7. Evolution of Google’s LLMs Transformer 2017 BERT (110M) 2018 T5 (220M) 2019 PaLM (540 B) 2022 Gemini (1.6 Trillion) 2023 In LLMs, parameters are the weights and biases learned during training, enabling the model to process input, make predictions, and generate outputs based on patterns in data.
8. Gemini API Capabilities • Supports multimodal inputs: text, images, audio, video. • Features like function calling, JSON mode, and embeddings. • Compatibility with OpenAI and many other open-source libraries for seamless integration.
9. Why Connect GenAI Models Externally? • Access to real-time and domain-specific data to prevent your GenAI model from hallucination • Integration into enterprise workflows and applications. • Enhancement of user interactions through dynamic content generation. • Fine-tuning is expensive and resource-intensive
11. Why Connect GenAI Models Externally? • Access to real-time and domain-specific data. • Integration into enterprise workflows and applications. • Enhancement of user interactions through dynamic content generation.
12. Methods of Integration • API-Based Integration • Function Calling • Embedding in Applications
13. Google Cloud Tools for Connecting Gemini • Google Cloud Functions • Google Cloud Pub/Sub • BigQuery
14. Overcoming Common Challenges • Latency and Performance • Security and Compliance • Scalability
15. Best Practices for Integration • Efficient API Usage • Data Validation • Monitoring and Logging
16. Responsible AI Integration • Bias Mitigation • User Privacy • Transparency
17. Thank you Lahore