Suggestions
Ignas Mociunas
Data & AI Solutions | Data Governance | GenAI & MLOps | B2B Platforms
Ignas Mociunas is a board member of ZüriLit, an organization that connects Lithuanian professionals in Zürich. He has notable accomplishments in the field of artificial intelligence, currently advising on AI-related projects.
In addition to his role at ZüriLit, Mociunas is recognized as a data specialist and senior consultant with extensive experience in reporting and data analysis. His professional background includes a variety of roles that leverage his expertise in data-driven decision-making and technology.12
Highlights
If you're looking for hands-on examples of 𝗔𝗜 𝗖𝗼𝗽𝗶𝗹𝗼𝘁 implementations, here's a great resource (not mine!) that could be helpful to reverse-engineer or clone👇
𝗪𝗵𝗮𝘁 𝗶𝘀 𝗶𝘁 𝗮𝗯𝗼𝘂𝘁? 👉 𝗥𝗲𝘁𝗮𝗶𝗹 𝗔𝗜 𝗖𝗼𝗽𝗶𝗹𝗼𝘁 𝗳𝗼𝗿 𝗖𝘂𝘀𝘁𝗼𝗺𝗲𝗿 𝗦𝘂𝗽𝗽𝗼𝗿𝘁 🤖 Git repo with 2.2k forks 🍴 and 339 stars ⭐
𝗪𝗵𝗮𝘁'𝘀 𝘁𝗵𝗲 𝘁𝗲𝗰𝗵? 👉 Azure 𝗔𝗜 𝗦𝗲𝗮𝗿𝗰𝗵: to create and manage search indexes for product catalog data 👉 𝗖𝗼𝘀𝗺𝗼𝘀 𝗗𝗕: to store and manage customer purchase history data 👉 𝗢𝗽𝗲𝗻𝗔𝗜: to deploy and manage the core models required for RAG-based architecture 👉 RAG pattern / Prompty asset / Promptflow frameworks 🧠 👉 Prompts for GPT-3.5-turbo (chat completion) and GPT-4 (chat evaluation)
𝗪𝗵𝗮𝘁'𝘀 𝗶𝗻𝘀𝗶𝗱𝗲? 🔧 Sample model cfg, chat & evaluation prompts for a RAG-based copilot app 💬 Prompty assets to simplify prompt creation & iteration for copilot scenario 🛒 Sample product and customer data for the retail copilot scenario 🤓 Sample application code for copilot chat and evaluation workflows ☁️ Sample azd-template cfg for managing the application on Azure
Here you go: https://lnkd.in/eJGxGFhZ
#handson #LLM #LLMOps #deployment #azure #GenAI #dataengineering #end2end #chatbot #retail
Last week, I was lucky to get a last-minute pass for 𝗕𝘂𝗶𝗹𝗱 𝗔𝗜 𝗗𝗮𝘆𝘀 by Microsoft.
The conference was happening around the corner in Zürich 🇨🇭 so there was no excuse not to join, I'm glad I attended!
Here's what you missed:
🚅 Keynotes by Catrin Hinkel and Judson Althoff sharing the market outlook, experiences of 𝗖𝗹𝗼𝘂𝗱 👉 𝗗𝗶𝗴𝗶𝘁𝗮𝗹 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻 👉 𝗔𝗜 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻, product portfolio to support the transitions, previews of #copilot use-cases, capabilities and numerous partner success stories. 👏
I was curious to see how #azure compares to other cloud providers my teams used in the past in the ML/DS/AI context; I admit the extensive integration of copilot makes it very appealing.
🔧 Workshop 𝗙𝗿𝗼𝗺 𝗠𝗟 𝗢𝗽𝘀 𝘁𝗼 𝗚𝗲𝗻𝗔𝗜 𝗢𝗽𝘀 by Hilda Kosorus from Crayon gave a great overview of ML deployment flows, showing extra layers of complexity coming with GenAI deployment practices as well as metrics we could use to evaluate the results. 📋
🔧 Workshop 𝗨𝗫 𝗣𝗮𝘁𝘁𝗲𝗿𝗻𝘀 𝗳𝗼𝗿 𝗟𝗟𝗠 𝗔𝗽𝗽𝘀 by Sevgi Uzungelis and Juan Manuel Servera Bondroit covered the best #design practices & toolkits, some good advice (e.g. add friction, avoid humanizing AI) to manage user expectations. Did you know there are #react component libraries for LLM Apps? 🎨
🔧 Workshop 𝗔𝗱𝘃𝗮𝗻𝗰𝗲𝗱 𝗔𝗜 𝗔𝗽𝗽 𝗣𝗮𝘁𝘁𝗲𝗿𝗻𝘀 by Ophélie Le Mentec and Sascha Corti sharing solution architecture in Lufthansa implementation of customer service AI agents. Interesting problem of #metrics: using LLM to check the outcome of LLM, identifying a sensible evaluation, benchmarks & quality 🧠
🔧 Workshop 𝗣𝗿𝗼𝗺𝗽𝘁 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝘁𝗼 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗟𝗟𝗠 𝗢𝗽𝘀 by Wojtek Bak explained AI adoption cycle, need for a new Operating Model, transition from MLOps to LLMOps, LLM Lifecycles, and solution architecture for LLM evaluation, super interesting! 🤔 💡
𝗧𝗮𝗸𝗲𝗮𝘄𝗮𝘆𝘀: 👴
- The future ahead is interesting: we're in the very early stages of #research, the state-of-the-art hasn't been defined yet. That's both exciting and scary.
- #Experimentation is the necessary next step, it will take time & cost money, iterating on it is a prerequisite to moving forward and becoming tech leaders.
- #Data #Governance will be an even bigger headache 🤯 than ever. It has been hard enough, yet #GenAI brings additional planes of complexity that need to be handled.
- #Quality and #reliability are getting new dimensions: how do we write unit tests for prompts? CI/CD needs to be extended from software code to data schemas & queries, then to prompts and finally (for now) LLMs. Whose job should that be? 🤔
- #datateams will be growing, blending with engineering/devops, adding new roles as organizations mature.
More on this later...
P.S. Hello to some familiar faces GianPiero Pavesi Dr. Michael Huber 👋
#ai #conference #llms #datagovernance #datastrategy #transformation #machinelearning #datascience #zurich
