Syllabus, Exercises and Repository
Syllabus, Exercises and Repository
Embedded Files
Pre-requisite
Pre-requisite
To build the Agentic AI application, the following system prerequisites are required:
A minimum of 100GB of free storage space on your PC.
At least 8GB of RAM.
A stable internet connection for software updates and pull LLM model using docker.
Operating system compatibility: Windows 10, MacOS Mojave, or later, or a recent Linux distribution.
Python 3.10 or higher installed version for running the application scripts.
Docker installed for containerization and easy deployment of application components.
Chapter 1: Foundation and Coding Skills
Chapter 1: Foundation and Coding Skills
Foundation
Overview of Artificial Intelligence (AI) and Machine Learning (ML)
Evolution of Language Models: From Rule-Based Systems to LLMs (GPT, BERT, etc.)
Introduction to Retrieval-Augmented Generation (RAG)
What is Agentic AI? Autonomous systems and their role in AI
Applications of RAG and Agentic AI in real-world scenarios
Coding Skills
Setting up Python environment (Anaconda, Jupyter, VS Code)
Basic Python libraries for AI: NumPy, Pandas, Matplotlib
Chapter 2: Foundations of Natural Language Processing (NLP)
Chapter 2: Foundations of Natural Language Processing (NLP)
Foundation
Basics of NLP: Tokenization, Stemming, Lemmatization
Word Embeddings: Word2Vec, GloVe, and Transformers
Introduction to Transformer Architecture (Attention Mechanism)
Coding Skills
Implementing tokenization and embeddings using Hugging Face Transformers
Fine-tuning a pre-trained model on a custom dataset
Chapter 3: Building Agentic AI Systems
Chapter 3: Building Agentic AI Systems
Foundation
Autonomous decision-making and task execution
Key components: Perception, Planning, Action, and Learning
Examples of Agentic AI: Autonomous Agents, Multi-Agent Systems
Integrating RAG with Agentic AI for enhanced decision-making
Coding Skills
Building a simple autonomous agent using Python
Integrating RAG with an agent for dynamic knowledge retrieval
Chapter 4: Retrieval-Augmented Generation (RAG)
Chapter 4: Retrieval-Augmented Generation (RAG)
Foundation
What is RAG? Combining retrieval and generation for better AI outputs
Components of RAG: Retriever, Generator, and Knowledge Source
How RAG improves over traditional language models
Question Answering, Chatbots, and Knowledge Management
Use cases
Implementing a basic RAG pipeline using Hugging Face and FAISS (Facebook AI Similarity Search)
Building a simple question-answering system with RAG
Chapter 5: Building RAG AI Applications
Chapter 5: Building RAG AI Applications
Foundation
RAG with cohere
RAG with Llama3.1
RAG autonomous
Coding Skills
Implementing Agentic AI systems
Chapter 6: RAG Techniques and Optimization
Chapter 6: RAG Techniques and Optimization
Foundation
Improving retrieval efficiency: Dense Retrieval vs. Sparse Retrieval
Fine-tuning retrievers and generators for domain-specific tasks
Evaluating RAG systems: Metrics and benchmarks
Coding Skills
Implementing dense retrieval using Sentence Transformers
Optimizing RAG pipelines for speed and accuracy
Chapter 7: Real-World Applications and Capstone Project
Chapter 7: Real-World Applications and Capstone Project
Foundation
Future of RAG and Agentic AI: Multimodal RAG, Reinforcement Learning, and Beyond
Ethical considerations in deploying RAG and Agentic AI systems
Coding Skills
End-to-end project: Building a domain-specific RAG application
Deploying the application using Flask or FastAPI
Case Study 1: Building Agent AI in Investment Decision
Case Study 2: Agentic AI in Business: Meeting Agenda and Summary
Case Study 3: Agent AI in Data Analysis and Visualisation
Case Study 4: RAG for Legal Document Analysis
Software Tools
Software Tools
Agent AI + RAG
Agent AI + RAG
The diagram flow encapsulates the core processes of Agentic AI and RAG, seamlessly integrating data preprocessing, real-time retrieval, and actionable insights to drive decision-making. Each phase aligns with masterclass modules, providing participants with a comprehensive learning experience.
Data Preprocessing
Stage: Data chunking and indexing prepare diverse inputs for efficient retrieval.
Module: Logic Theory & Installation covers foundational techniques and system setup for scalable data processing.
Retrieval and Augmentation
Stage: RAG retrieves relevant data, enriched with real-time insights for dynamic outputs.
Module: Implementation & Use Case Development focuses on designing retrieval pipelines tailored to industry applications.
Actionable Insights
Stage: Predictive analytics transforms processed data into strategic decision-making tools.
Module: Practical Applications & Predictive Analytics teaches end-to-end workflows for forecasting and advanced data-driven solutions.
Evaluation and Refinement
Stage: Feedback loops ensure system accuracy and adaptability over time.
Module: Finalisation & Presentation enables participants to refine and showcase their projects with expert guidance.
This streamlined framework bridges theoretical understanding and practical execution, equipping participants with both technical expertise and strategic vision for deploying AI-powered solutions effectively.
Client Testimony
Client Testimony
New achievement unlock by Rendi Risandy (Alumni Batch 1)
Tax, Accounting, Transfer Pricing, Disputes and Technology enthusiast
In the last 3 week I have a chance to take a course about Agentic AI and RAG in Patria & Co. Before this course, I don't have any idea what is the work with AI and how to used. Therefore, this training is take me from the scratch to learning something new that I don't know before. As previously I still only learning Diploma in Tax Technology with Tolley and CIOT, as this is more learning about database and process with the system. It might be good if in the future this might be combine with AI.
As the end of the session with Pak Harry and team, we need to delivered and present the "product", what is the implementation of AI application to do our job better.
As at the beginning, my application is not working, and with the help of the team of Patria & Co especially Mas Valdi and Mas Hilmy, currently the agentic AI of my taxlaw searching is accomplished.
How this application work, Qdrant is working as our databased in our brain as Hippocampus that will generate by the Amigdala that will be generate by Phyton and Streamlit. And as the Cerebrum is perform by Open AI as the to generate the work. In tis sample, the machine will search to setpp that they will crawl to the site to take the information, even though it will take a long time to process, but after search one time, the information will store in the machine as the part of machine learning of the application. Therefore, the more the application is used, the more is databased wider. See also the sample of the coding below that contain only 1,000 rows. It can create more even complicated in the future.
My grateful to Pak Dr. Harry Patria for the help during the course for the explanation. You can check on the below picture as the sample of the result, using the menu and the output. This is still only prototype that only doing during after offices hour within 1 weeks after the offline meeting. In the future, AI might help your job if you know how to used it. And I have no regret to take this course.
Once again, I don't have the background of IT or AI and I'm the type of a millennial that might be "gaptek" with technology. As I always remembered the quote, "the magic you are looking for, is in the work you are avoiding".
Finally, as a quote from martin Luther king, "I have a dream", as my dream in the future might be the tax disputes in related to documentation with tax offices is reduce not only in Indonesia but all over the world because of the help of AI.
As a tax professional, don't worry to be replace by AI, we can work together and I hope AI can help us to be better in the future as a tax and legal professional. Stay learning to be the best version of yourself.
Prediksi Risiko Kebangkrutan Perusahaan dengan Machine Learning dan Agent AI by Sopian Hadianto (BPKH - Alumni Batch 2)
Bagaimana jika analisis risiko kebangkrutan yang biasanya memakan waktu berminggu-minggu bisa selesai dalam hitungan detik?
Dengan bangga saya membagikan sebuah proyek inovatif hasil "berguru" di Patria & Co. Platform Prediksi Kebangkrutan Perusahaan berbasis AI.
Platform ini bukan sekadar dashboard biasa. Di baliknya, kami mengimplementasikan arsitektur canggih yang menggabungkan:
► Agentic AI: Sebuah "agen" cerdas yang secara mandiri mengambil data, menjalankan model, dan menyusun analisis.
► RAG (Retrieval-Augmented Generation): Model tidak hanya menebak, tetapi "membaca" data finansial real-time dari Yahoo Finance untuk memberikan konteks yang kaya.
► Time Series Forecasting & XAI: Menganalisis tren historis dan memberikan transparansi "mengapa" di balik setiap prediksi.
Cukup masukkan ticker saham, dan AI Agent kami akan langsung bekerja untuk Anda:
> Mengambil 14+ rasio keuangan terbaru.
> Menganalisisnya dengan model Ensemble Classifier.
> Menghasilkan ringkasan analitik yang mudah dipahami.
Hal ini adalah perpaduan antara AI Generatif, Machine Learning, dan Desain yang berpusat pada pengguna. Tertarik mencoba? Anda bisa mengaksesnya secara gratis.
cc: Pak Dr. Harry Patria
Page updated
Google Sites
Report abuse