I you increase the code, hard to see. Very small

Tried to sign up to the Databricks site. Horrible access, i need to solve 10 cubes questions. Horrible idea

Good stuff from yaron haviv

They use GPT-3.5 for Lilli, very interessting. With RAG?

18:35 did they fine tune a model with company data?

CNC Miling Copilot is awesome. But how they do it. Finetuning?

Awesome video. Underrated. He should do more lectures

Why adds in this video??

subtitles please

Hey guys great talk!!! Any chance to get the slides? Would it be possible to do an executive summary in the beginning?

🎯 Key points for quick navigation: 00:00:00 *🤝 Introduction and Session Overview* - Introduction of the speakers and agenda. - Explanation of the webinar format and how to interact. - Overview of what will be covered: scalability, effective, and responsible building of customer-facing gen AI applications. 00:02:33 *💡 Setting the Context for Gen AI Applications* - The evolution of AI and its impact on businesses. - The importance of choosing the right platform and technology for gen AI applications. - Emphasizing the role of flexible and scalable environments. 08:05:00 *🔄 Unified Data Platforms for AI* - Discussion on the need for a unified data platform. - Importance of storing and accessing complex data types. - Flexibility in connecting various applications, frameworks, and clouds. 13:10:00 *🚀 Beginning the Gen AI Journey* - Poll to understand audience's stage in their gen AI journey. - Examples of successful implementations and potential pitfalls. - Addressing the need for governance, cost control, and risk management. 18:22:00 *📈 Gradual Approach to Customer-Facing Apps* - Step-by-step approach for implementing gen AI in customer-facing applications. - Detailed stages of from call analysis to chatbots. - Case studies and real-world examples of gradual implementation. 22:03:00 *🛠️ Building Effective AI Pipelines* - Breakdown of complex AI application pipelines. - Integration of traditional AI with newer gen AI technologies. - Use cases emphasizing the importance of comprehensive and well-structured AI pipelines. 00:24:21 *🏭 Synchronizing AI Factory Processes* - Overview of multi-dimensional CI/CD process: Synchronizes software, models, and datasets, - Importance of monitoring models for performance, data drift, and possible security risks, - Continuous operation feedback loop ensures adapting to fast changes and risk mitigation. 00:26:14 *🧠 Building an Intelligent Chatbot* - Introduction to the concept and architecture of a smart credit card chatbot, - Utilizing hyper-personalization by fetching financial and personal data, - Query refinement, session loading, and context preservation to streamline responses. 00:30:06 *🔄 Personalization in Chatbots* - Chatbot adapts its tone based on client's age, using a casual tone for younger clients and a formal tone for older clients, - Selection of the best credit card offers tailored to client's preferences and financial history, - Continuous dialogue with emojis and cheerful responses for younger clients versus respectful tones for older clients. 00:35:37 *📊 Data and Algorithm Utilization* - Displaying synthetic client data and credit card details used to personalize recommendations, - Algorithm's ability to match clients with appropriate credit cards based on minimized fees and targeted perks, - Importance of data integrity and management for continuous improvement of models. 00:37:03 *🚀 Starting with Gen AI Applications* - Discussion on moving from experimentation to live applications with limited scale and risk, - Importance of building a foundational data and governance infrastructure for scaling AI applications, - Real-world examples of evolving Cloud and on-prem AI deployments based on client needs and regulatory requirements. 00:47:10 *🎁 Workshop Giveaway and Q&A* - Announcement of workshop giveaway for attendees to help address their specific AI challenges, - Transition to a Q&A session focusing on practical steps and considerations for implementing and scaling Gen AI applications, - Emphasis on the importance of governance, monitoring, and continuous improvement in AI projects. 00:49:56 *📈 Adoption of AI in Banking* - The rise in AI spending among top US banks is justified by expected business benefits. - Implementation time for AI projects varies greatly, with easier in-house projects potentially rolling out within weeks. - Security and data privacy are managed by MongoDB via TTL indexes, encryption, and other client-side security features. 00:52:44 *🛡️ Governance and Security in AI Applications* - Governance and guardrails around AI include monitoring, filtering, and using LLM as judges before producing answers. - Prompt injection and boundary issues are managed through regulatory compliance and personalization while ensuring data privacy. 00:55:53 *🏦 Handling PII and Compliance* - Managing personally identifiable information (PII) requires secure training and fine-tuning models with real client data. - Solutions for handling PII include in-cloud and on-prem deployments, allowing for GDPR compliance and secure data usage. - Enterprises can leverage existing client data to enhance AI applications without breaching regulations. 00:57:39 *💡 Common Challenges in Scaling AI* - Scaling AI applications often boils down to managing costs effectively, particularly in GPU usage. - There is a knowledge gap in how to optimize GPU resources, which can lead to exponentially high costs if not managed. - Proper education on GPU usage can lead to significant cost savings and improved application performance. 00:59:12 *🌐 Resources and Closing Remarks* - Participants are encouraged to explore additional resources, including blogs, case studies, and previous webinar sessions available on the website. - Gratitude is expressed to the presenters and participants for their insightful contributions and questions.

Gen AI is transforming the financial industry, with use cases in virtual experts, content generation, customer engagement, and coding acceleration. However, successful implementation requires addressing risks, building the right operating model, and scaling the solutions. Key areas driving adoption include risk management, productivity gains, and organizational change management. Key moments: 00:26 The webinar discusses challenges in implementing AI, especially in financial services, and highlights how leading organizations are successfully using AI in their business environments. Larry and Yon will delve into specific challenges and solutions in AI implementation. -Transition from analytical AI to generalized AI is transforming enterprises, showcasing the power of analytics. Companies are expanding into AI beyond POCs. -Gen can add significant value to the banking sector globally, with themes like risk management emerging as crucial. Enterprises face challenges in scaling AI from POCs to production environments. -Gen is amplifying the value analytics can deliver to the banking sector, but it's not suitable for all situations. Various use cases show the versatility of gen in different industries. 08:04 Efficiency gains are driving 75-80% of the value in the office today. Banks are experimenting with virtual experts, content generation, customer engagement tools, and coding acceleration to increase productivity and efficiency. -Banks are exploring virtual experts to boost productivity through internal and external access, such as to customers. -Content generation tools are aiding in producing initial drafts of contracts, reports, and legal documents, requiring human oversight for finalization. -Customer engagement tools like chatbots and co-pilots are enhancing interactions between bankers and customers, improving service operations. 16:07 The video discusses the implementation of new technologies in banking, such as a 20% increase in customer service efficiency and the importance of experience in coding productivity. It also highlights the development of a knowledge agent tool by McKenzie for professionals. -The importance of experience in coding productivity, focusing on software developers with the right level of experience for efficiency gains. -McKenzie's development of a knowledge agent tool with access to extensive knowledge documents, financial data, and expert interviews for professionals. -The challenges of deploying new capabilities at scale, emphasizing the strategic positioning within the enterprise, data architecture, and process changes. 24:10 Moving from prototyping to production involves building more seriously, managing risks, and increasing accuracy. This transition requires three pipelines: data, testing models, and application, along with monitoring and governance for quality assurance. -Data pipeline involves managing document ingestion, content indexing, metadata, and model testing for accuracy and risk management. -Testing models includes evaluating, deploying, and monitoring models to ensure accuracy, performance, and controlled upgrades. -Application pipeline encompasses data enrichment, model execution, governance, and post-processing, with a focus on telemetry data for continuous improvement. 32:13 The video discusses the process of data pipeline in an application, including data ingestion, filtering, duplication, privacy protection, tokenization, and indexing in a vector database, to ensure accurate and secure data processing. -Classification mechanism challenges in understanding user queries and composing holistic questions from session history for accurate data processing. -Data refinement with history, classification of user queries, and selection of processing paths based on user intent for effective data retrieval and response generation. -Importance of testing, monitoring, and continuous improvement in application pipelines to maintain data quality, prevent risks, and ensure performance through version changes. 40:17 Uber's revenue data is indexed and stored in a vector store, combined with llm information to generate answers. The process involves refining questions using historical data and current queries to improve results. -Use of llm models like open AI for refining questions and generating answers based on historical and current data. -Incorporating multiple models such as Vision, Audio, and Transcription models in applications for analyzing client calls and generating insights. 48:21 Access to the internal AI tool Lily is restricted to within the organization, but its success has led to collaboration with external institutions for knowledge sharing and development. Technologies used in the application include customizable databases, ML models, and cloud services for flexible and efficient data processing. -Collaboration with external institutions for knowledge sharing and development based on the success of the internal AI tool Lily. -Technologies used in the application, such as customizable databases, ML models, and cloud services, for flexible and efficient data processing. -Challenges and strategies related to talent acquisition and upskilling in the financial industry to meet the demands of evolving technologies like AI and data science. 56:25 McKenzie typically ensures client involvement in the build process, promoting learning and eventual client ownership of applications post-engagement, fostering a collaborative and educational approach. -McKenzie's approach to client engagement and knowledge transfer during the build process, emphasizing client ownership post-engagement. -Support for multiple languages in projects, including strategies like translation and model fine-tuning for proficiency. -Challenges and considerations in AI model accuracy, data privacy, bias, and mitigating risks in AI applications.

Very insightful. Really love to learn these things which you had shared with us. Love to learn the fashion project.

Looks super interesting! So many questions! I skimmed a bit the docs but I could find all answers, maybe I need to look closer! How are transformers available? Seems like it may be based on some base docker image that is somehow configured in that yaml file for the function? What if I need to import something else that is not available on the base image? (how would I "pip install" etc.) How about functions needing arbitrary assets? Nuclio seems to use no DB, where is the data / configuration stored? k8s CRDs?

Not sure why McKinsey just makes up acronyms instead of using generally accepted nomenclature. Seems arrogant and pretentious, let alone confusing.

hi, can you share the notebook please?

put link to app please

This is a great presentation of GenAI overview. Thanks.

I am forever baffled why and how such videos have so few views and comments, but a random tiktoker eating corn or whatever has millions of reactions. Thank you for posting this.

Very Insightful video, Thanks for sharing your knowledge.

Please, can you share link to GIT?

Why no notebook in description?

Superb ✨

Thank you for sharing this good info. Teach me more about AI

I would like to hire you all to develop my app.

Very Insightful, thank you for sharing this wonderful presentation.

Where's the link so I can do this? I have a business plan for an AI companion app, but I don't know this coding stuff nor where to go.

I'm using mlrun with docker.

I'm getting error in invoke function step. Can u please help regarding that.

How to create a docker node? can u please help me with that?

Thanks for the tutorial, I installed the MLRun but the functions disappear with the docker desktop restart, any idea why?

Is that windows 7 I see in 2022?

Can you upload the same for vscode

Nice demo! The model is from s3. How should I set up some on-premise storage for model and deploy it with proper version control?

Any advantage comparing to MLflow?

I hope that more relevant data scientists will see this video and will start thinking about it.

𝚙𝚛𝚘𝚖𝚘𝚜𝚖

great tutorial sir thank you god bless

Wow video

Q: if the data from the Kafka streams are needed to process an inference request, how do you guarantee to have the data in the feature store before processing the inference request?

Thanks a lot team @Iguazio. Very much useful content. How to identify a kind of drift in a model and when to go for model archiving or re-placing a model instead of existed one and why? Thanks in advance.

Very useful!

Thank you for this video. Very informative.

Too much technical terms. Can you do a detailed video on mlrun with an example

Great event!

very interesting discussion, thank you for hosting it!

Ххх

No sound?

I'm sure it wouldn't be hard to customize, but just curious -- does the canary deployment system here support multiple canary models or just one at a time?

Екатерина купила крым заплачено у хрущева крыша поехала подарить крым а реФерендума россиян не было крым законно России принадлежит байден мечтать не вредно.