Learn how to build your own open-source AI software for construction calculations
- 1. Eliminating rework and design errors on construction projects
- 2. Topics of this presentation 1. High level intro to topic of AI 2. What are AI agents? 3. How Are Civils.ai using AI agents? 4. Using Civils.ai vs building your own AI agents?
- 3. The simplest explanation of AI/ML Known input Known output Unknown function
- 4. A quick history of AI, it’s older than you think ● AI algorithms are are mathematical constructs (domains, functions, transformations) ● Initial foundations of artificial neural networks were laid by Gauss and Legendre (~1805) ● In 1950s we started mixing simulation/heuristics with regression but no computing power! ● We deviated to simulation/heuristics and left ANNs on the curb ● 1986 Backpropagation! ○ 1998 LeNet! ■ 2012 AlexNet! ● 2022 Chat GPT
- 5. Why is AI becoming increasingly powerful? The number of transistors on a chip doubles every 2 years ● GPUs are becoming cheaper, faster and resource efficient ● New programming libraries are being developed: ○ Tensorflow ○ PyTorch
- 6. What do we mean by data science, AI & machine learning? ● Data science is overall subject (typically the subject you’d study) ● Artificial Intelligence is any non- natural construct which can take a decision without human intervention ● Machine learning is ONE approach to create AI algorithms ○ Others? Rule based system (if, else), bayesian systems etc. ● ANNs are generic function approximators ● Deep learning refers to any Artificial neural network (ANN) network which is more than 2 layers! Neural Networks Deep learning
- 7. Most us are already using foundations of ML in Geotech ● Regression is a commonly used method in machine learning ● Many empirical formulas we use in our calculations are based upon some form of regression ● But how is it best to calculate these formula?
- 8. Different technologies can be used for the function approximation Classical Machine Learning (Most AI apps use) Deep Neural Network (Text to speech) Convolutional Neural Network (Vision intelligence) Transformer Model (Large Language Models) Recurrent Neural Network (Text to speech) Graph Neural Network (Recommendations) Autoencoder (Recommendations) Generative Model (Image generation)
- 9. Introduction to concept of AI Agents
- 10. What are transformer models? ● A transformer model is a neural network that learns context (and meaning) by tracking relationships in sequential data like the words in this sentence. ● Foundational models you may have heard of: ○ GPT (OpenAI) ○ BERT (Google) ○ LLama 2 (Meta) ○ Mistral ● Not just Large Language Models
- 11. AI Agents can be built by abstracting upon foundational models ● AI Agents are systems or software capable of autonomous action in order to meet specific objectives without the need for continuous human guidance. ● These agents can learn, make decisions, and take actions that maximize their chances of achieving their goals. ● Can be chained together to create AI workflows ● All AI agents discussed in this presentation are some form of Transformer Model
- 12. An example AI Agent configuration ● One agent does the work, another checks the output, resulting in improved accuracy and reliability at completing user objectives ● In this case an assistant agent refines and guides a ‘User Proxy’ agent in executing python scripts to arrive at the objective ● Without the assistant agent (or human intervention) the ‘User Proxy’ would have no feedback that it is headed towards the objective
- 13. There are multiple AI Agent configurations They can be good at different types of task and there are always new configurations or ideas emerging
- 14. The power of AI agents to solve complex problems Challenge: Detect who’s not wearing a helmet?
- 15. Building your own AI tools vs Civils.ai Building your own: ● Building an MVP isn’t hard. Building something which works is hard! ○ If you have a talented developer to start setting something up, are they always going to be around to maintain the system? ○ If you don’t have a developer then challenges around cloud management (GPU servers are expensive), building AI pipelines, fine- tuning / training custom models. (hard problems) Using Civils.ai: ● Can host data following your requirements (run on your company server) ● Our AI models are geared towards construction with proven results (established commercial relationships with multiple construction orgs). ● We can fine-tune our existing solution to help enable pilot projects, no custom IT project is required. ● We have API’s to help you connect outputs your own dashboards and in- house tools.
- 16. The best places to start are with LangChain & Autogen libraries AUTOGEN is another github.com/microsoft/autogen LangChain is popular github.com/langchain-ai/langchain
- 17. Project idea - Create a chatbot to explain specific documents Use RAG (retrieval augmented generation) to give an AI Agent some specific knowledge Query Results
- 18. We provide training with all our licenses Our software licenses come with 4.5 hours of practical demonstrations in AI fundamentals, strategies for implementing AI & even building your own in-house. - Lesson 1: Course introduction - Lesson 2: Foundations of AI - Lesson 3: Implementing AI with no-code (like CIvils.ai) - Lesson 4: Building AI with python - Lesson 5: Publishing your AI applications - Lesson 6: Fine tuning AI to construction - Lesson 7: AI regulations and safety Available on www.civils.ai + Free educational content on
- 19. Thank you for listening stevan@civils.ai