GPT in the company: Develop your customized Large Language Model
.svg)
.svg)
.svg)
Machine Learning Engineer
Basics
- Introduction to AI and Large Language Models (LLM)
- Relevance of LLMs for companies
- Potential use cases and their added value
- Product development: technology, best practices, architecture
Application examples from practice
- Bosch and Aleph Alpha
- DM and dmGPT
- Further examples: Moody's Copilot, OTTO ogGPT, McKinsey Lilly
- Discussion and analysis of the applications presented
- Types of in-house GPT: enterprise search, process automation with prompt engineering, products with AI
Approaches and possibilities
- Training from scratch: advantages, disadvantages and costs
- Fine-tuning: adaptation to internal company data, quantity of training data, quality
- In-context learning: AI's ability to recall context
- Retrieval Augmented Generation (RAG): Integration of databases, vector store, customizing
InhouseGPT Pilot
- Process: Prototype with ChatGPT, pilot with Streamlit & Chainlit, InhouseGPT as a scalable product
- Extraction of company data from internal company data
- Knowledge base with Obsidian and Markdown
- Adding data to qdrant as vector embeddings
- LLM: Use of OpenAI interfaces or setting up Llama 3.1.405B
- Prompt development and evaluation with Python
- Improvement of the RAG pipeline with hybrid search and reranking
- Deployment with Github and Streamlit Share
Technologies and tools
- Selection of suitable infrastructure (e.g. cloud vs. on-premise)
- Models: OpenAI API and GPT-4o, Claude Sonnet 3.5, Llama 3.1 405B
- Data: qdrant, Weaviate
- Orchestration: LangChain, LlamaIndex
- Interfaces: Streamlit & Chainlit, Business Apps, Web Apps
- Deployment: local, interface, cloud VM
Staffing and implementation
- Requirements and expectations of the stakeholders
- Team structures and roles
- Requirements for qualifications, skills and further training
- Development of a solution architecture; costs and budget
- Training and empowerment of end users
Important note on the pre-installation of software
The training includes many practical exercises to test the techniques described. You will carry these out on your own computer. Please ensure that Python 3.12 and Git Access Token are installed on your computer before the seminar. Furthermore, Python packages will be installed during the training using the package manager "pip packages". Please try this out before the seminar date and contact your IT administration if you have any problems.
- You will acquire comprehensive knowledge about the development and application of large language models in a corporate context
- You will learn how to adapt LLMs to your specific business requirements
- You will learn how you can set up an intelligent search by using LLMs
- You know what you still need to implement your own GPT and what steps lie ahead of you
The content of this training supports the obligation to provide evidence of the promotion of AI competence within the meaning of Art. 4 EU AI Regulation.
This training training is conducted in a group of a maximum of 12 participants using the Zoom video conferencing software. Individual support from the trainers is guaranteed.
The practical exercises are carried out together in the group. Python, Git, Obisidian and Cursor will be used.
The trainers are on hand to answer your questions - in the virtual classroom or individually in break-out sessions.
Once you have registered, you will find all the information, downloads and extra services for this training course in your online learning environment.
This training is suitable for developers in the field of artificial intelligence and machine learning as well as specialists who are involved in the implementation of AI solutions in companies.
.svg)
