Code Mania 1011 in Nutshell

By Charunthon Limseelo (boatchrnthn) and Collaborators

Greetings.

First of all, I haven’t joined much event that gives us some tech inspirations because I always come in tech annoucement and meetup more than tech life-wellbeing meetup. However, things a bit change by now since I got nothing to do for my semester break and bored in the house. So, I came on this tech meetup for helping Thai Programmer Association in the event name “Code Mania 1011”, which is the 10th anniversary of Code Mania, hosted by Pongsiri Pisutakarathada (P’ Save Pongsiri), and over 250+ attendees join our event on that day.

In this meetup event, there would be no annoucement of new features of big tech companies, but they would like to show how AI goes beyond in nowadays, even on office tasks or our daily life. Then, I’m going to pick some interesting points that I would like to share to you from this event.

Next Generation AI for Developers — Microsoft MVP

As we embark on this journey, we look forward to a future where challenges once deemed insurmountable become opportunities for groundbreaking innovation, ushering in an era where the difficult or impossible is made possible. — Teerasej Jiraphatchandej

In the realm of app development, AI is not only transforming our workflows but also enabling the creation of entirely novel applications that were previously unimaginable. Tools like GitHub Copilot are pivotal in this transformation, offering coding assistance that accelerates development cycles and enhances collaborative problem-solving.

GitHub Copilot and Copilot Workspaces, as AI-powered coding assistants, are the good examples that revolutionize how developers interact with code. By suggesting code snippets in real-time, Copilot reduces the time and effort required for coding tasks, thereby allowing developers to focus more on innovation and teamwork. This tool is a cornerstone in the evolution of developer productivity and efficiency.

Moreover, AI models such as GPT-4o, available on Azure OpenAI, expand the horizons of application development by offering robust capabilities and integrations within Azure AI frameworks. These models, ranging from comprehensive language models to specialized vision models, empower developers to build sophisticated applications efficiently across multiple programming languages.

In the realm of AI services, Azure AI Studio exemplifies the future of application development, providing a platform for testing and evaluating AI models seamlessly through Phi-3 Serverless APIs. This integrated approach fosters rapid prototyping and deployment of AI-driven solutions, pushing the boundaries of what developers can achieve today and tomorrow.

Embracing Responsible AI Innovation, our initiatives prioritize built-in security measures and risk evaluation frameworks, ensuring that AI technologies are deployed safely and ethically. By mitigating potential risks, we pave the way for sustainable AI-driven innovation that benefits society as a whole.

Software Engineering Beyond the AI Hype — CreatorsGarten

Image from K.Pongsiri Pisutakarathada

Let’s remember the wisdom of Dr. Alan J. Perlis: “A language that doesn’t affect the way you think about programming, is not worth knowing.” And the words of Silpa Bhirasri remind us: “If you don’t read books, what will you know?” — Phoomparin Mano

Four years ago, the buzz was all about cryptocurrency and blockchain. Should we embrace blockchain’s decentralized, transparent, and immutable aspects? Or opt for centralization, privacy, and scalability? It seemed like every discussion circled back to these questions, echoing the “law of the instrument”: Are we now swapping crypto fervor for AI obsession?

Navigating the Hype Cycle of emerging technologies has been a journey in itself, peeking at lofty expectations and deciphering what truly defines the trend.

But let’s delve deeper: What are the natural and theoretical limits we face? From pondering the feasibility of building intergalactic starships to unraveling the mysteries of AI, these are questions that push the boundaries of our understanding.

In the realm of AI, we confront its probabilistic nature and the imperative to never use what we cannot verify. Optimization-based learning and generalization demand that we grasp problems thoroughly before attempting solutions. It’s not just about writing code; it’s about having a profound domain model to guide our efforts.

As we delve into active research areas like mechanistic interpretability and the works of pioneers like Neel Nanda and Penzai from Google, we strive for deeper alignment, understanding, and control in our technological pursuits. For junior engineers, the challenges lie not just in solving technical problems but in achieving product-market fit and mastering the essentials of software development.

TiDB in the era of GenAI — Wing Yao (PingCAP)

TiDB is a database system that makes use of the MySQL driver and scales exceedingly well, making it an excellent choice for handling heavy loads. It supports SQL and has recently added a Vector Search feature, which allows us to manage data using SQL commands as usual. This vector is a native type within the database, and TiDB now offers a serverless version for testing.

Embarking on an AI Application Journey

Building an AI application using TiDB can be an exciting journey. One approach involves using LLM (Language Model) to convert text queries into SQL commands, which then fetch data from GitHub Events stored in TiDB.

This idea of using Prompt Engineering can be applied to Text2SQL apps. By writing effective prompts, they can guide the LLM to produce more accurate answers. However, achieving high accuracy is still a challenge due to bottlenecks such as model parameters, few-shot learning techniques, token limitations, and hallucinations.

Overcoming Bottlenecks with RAG

To overcome these bottlenecks, all data including text, images, and videos should be converted into vectors first. Then, they can use a distance function to find similarities.

Applying the RAG (Retrieval-Augmented Generation) model can help improve accuracy significantly, from 51.48% to 75.46%. TiDB has utilized RAG in another app, the TiDB Q&A Bot. However, RAG also has limitations and tends to perform worse with larger and unrelated data.

Enhancing Knowledge with Graph Knowledge RAG

LLM can be used to create a knowledge graph, and RAG can be utilized in the process. The results obtained from these processes can then be compared and fine-tuned.

Essential Features for an AI Database

An AI database should have the capacity to store vast amounts of original data in various formats such as SQL, NoSQL, and Vector. It should support vector and semantic search and be capable of arranging relationships into a graph.

TiDB offers a solution that combines the complexities of data and database management into the database itself. This approach can, however, lead to vendor lock-in. Thus, the choice of database should be based on available resources such as budget, team skills, and specific use cases.

Data in the age of AI — Kittipong Ruksa

Image from K.Pongsiri Pisutakarathada

The path is not defined, but the solution starts with understanding and action. — Kittipong Ruksa

In this session, successful AI implementation requires meticulous planning, combining data and advanced technology to foster new innovations.

Generative AI remains a topic of curiosity and skepticism, especially among the Gen Z demographic. While AI assistants are in use, complete trust is not yet established. The future lies in AI-driven solutions, contingent upon high-quality data to ensure continuous improvement and optimal customer experiences, as depicted in the cyclical flow.

It’s imperative to safeguard customer ideas and data within the framework of Densu’s Generative AI Guideline, emphasizing robust enterprise data management. This includes evaluating business value, designing from current to desired states, comprehensively understanding and managing data across business domains, ensuring accountability, measurement, control, utilization, and enhancement.

Governance is pivotal, supported by a detailed Data Quality Framework, specifying ownership, domain-specific processes (e.g., customer versus financial data), and approval mechanisms.

The CX Technology Stack, initially puzzling, centers on enhancing Customer Experience (CX) through embedded AI models and ongoing digital transformation efforts, prioritizing rapid feedback loops for effective Tech-Biz alignment.

Use-case scenarios, like Mugen AI reducing costs and enhancing conversion rates, underscore the evolution from B2C to M2M interactions, guided by the Data Exchange Value Index for comprehensive applicability and understanding.

Building applications with generative AI entails a strategic approach to harnessing data’s transformative power, ensuring sustainable innovation and operational excellence.

Building Applications with Generative AI — borntoDev

Every time we use it, we must always consider the Alternative Flow or other options. Try to look from the user’s perspective. If our target is the general public… And we must make it as easy as possible for the general public to use. — Kittikorn Prasertsak

Today, AI is integrated into our daily tasks and applications, such as generating office and meeting summaries for end users. Developers benefit from tools like GitHub Copilot and JetBrains AI Assistant, which assist in coding, generating test cases, and refactoring code, while also providing insights into potential improvements.

These advancements are driven by generative AI, accessible not only to large corporations but also to startups and innovators. Examples include Castmagic, transforming chats, meetings, and podcasts into content, and Formula Bot, which simplifies tasks like transcribing video and audio into text or converting conversations into specific formats like Excel formulas, SQL queries, or App Scripts.

Afforai aids in managing academic papers by suggesting vocabulary, finding citations, and formatting content, enhancing research efficiency. FlutterFlow enables AI-driven app development by designing, coding, testing, and deploying applications based on user specifications.

BorntoDev focuses on practical AI applications, such as creating context-aware chatbots for video content, ensuring accuracy through active learning where human oversight ensures the AI’s responses align with expectations.

For those looking to develop AI chatbots independently, key considerations include understanding user pain points and motivations for app usage. In Thailand, apps like Line are immensely popular, with tools like Line OA integrating with AI solutions such as ChatGPT or Gemini, offering free courses for developers to explore and leverage AI effectively.

These initiatives highlight the transformative potential of generative AI in empowering developers and enhancing user experiences across various applications and industries.

Streamlining Data Extraction with Generative AI — Krungsri Nimble

Image from K. Wittawat Karpkrikaew

The best option may not always be the answer. Choose what is appropriate. — Sittisak Chumpreecha

At Krungsri Nimble, we understand the challenges customers face when transitioning to API-based solutions, especially when dealing with paper documents or PDFs. Leveraging AI in data extraction is crucial to avoid missing out on business opportunities.

Why Data Extraction is Important

• Streamlined Workflow: Traditionally handled by humans, AI enables immediate processing without scalability issues.
• Efficiency and Productivity: Ensures accurate data extraction, reducing time-to-market.
• Informed Decision Making: Rapid and accurate data availability facilitates quick decision-making.
• Cost Saving: Insights gained from data and increased automation reduce operational costs.

AI in Data Extraction

Krungsri Nimble employs Azure Document Intelligent (formerly Form Recognizer) for learning document templates and extracting data. Generative AI tackles document complexities, including data security challenges, ensuring robust handling of sensitive information.

Nimble’s AI Solution

• Data Accuracy: Ensures precise and reliable data extraction.
• Security: Complies with PDPA regulations and guarantees data integrity.
• Transaction Cost: Minimizes operational costs through efficient workflows.

Workflow Overview

1. Input: Handwritten documents, PDFs, scanned PDFs, or JSON files.

• Preprocessing: Classifies document types (e.g., handwritten, scanned PDF) for optimized processing.
• Text Extraction: Verifies document integrity and metadata classification.

2. Data Cleaning: Prepares data for Generative AI processing, ensuring consistency and eliminating anomalies.

3. Generative AI:

• Method Selection: Chooses appropriate methods such as chat or file search based on document type.
• Model Selection: Matches AI models to data characteristics (e.g., data tables use OCI models, handwritten notes use OCR).
• Prompt Preparation: Configures prompts to align with document formats:

1. Processing: Accounts for model limitations (e.g., token limits, cost considerations, output format uniformity in JSON).
2. Transformation & Verification:

• Transformation: Converts JSON data into usable objects for error detection.
• Verification: AI and human review ensure accuracy, with editors available for final adjustments to meet desired output formats (e.g., JSON, Excel).

In conclusion, before implementing AI, understanding its suitability and capabilities is crucial, especially with the continuous evolution of AI models. This ensures optimal integration and maximum benefit from AI-driven data extraction solutions.

Monetize your business with Bitkub AI — Bitkub

The Era of AI

AI’s journey began modestly in the 1950s with rudimentary if-else statements pioneered by Alan Turing and John McCarthy’s early concepts of Machine Learning/AI. Tools like ELIZA emerged for psychological therapy.

However, AI experienced periods of “winters” marked by overhyped expectations that didn’t align with reality due to funding and research constraints. Expert Systems and hardware limitations defined the challenges of the second AI winter.

Key milestones emerged, such as IBM Deep Blue in 1997, the convergence of ML and Big Data in 2000, the rise of Deep Learning in 2010, and breakthroughs like AlphaGo in 2016 and Generation AI like GPT in 2020, signaling a renewed era of AI innovation.

AI Renaissance in Today’s Era

Today, AI’s resurgence is underpinned by:

• Data: Improved quality and accessibility.
• Hardware: Advanced CPUs/GPUs/TPUs like NVIDIA Blackwell.
• Algorithms: Complex neural networks with adaptive functions, akin to the foundational if-else statements but significantly more sophisticated.

From the World Economic Forum’s perspective, successful AI adoption involves:

• Initial Costs: Investments in app downloads or proven technology deployments.
• Additional Costs: Secondary and tertiary user adoption.

Utilization Trends

AI plays dual roles:

• Augmentation: Enhances tasks such as customer service with ChatBots, freeing up employees for more strategic work like sales.
• Automation: Replaces human intervention, exemplified by tools like Figma.

According to Satya Nadella, Microsoft’s CEO, companies with over 50 AI-driven employees can achieve productivity levels comparable to those with 500 employees.

AI Infrastructure

AI’s framework comprises:

• Infrastructure Layer
• Model Layer
• Application/Agent Layer

For aspiring unicorns, effective AI deployment requires addressing:

• Individual Challenges: Skill gaps, time constraints, and cost implications of using multiple tools.
• Organizational Issues: Data security, integration with existing systems, and ethical considerations.

Bitkub’s Approach

Bitkub tackles these challenges by:

• Establishing an AI-focused team to study and resolve organizational pain points.
• Implementing a roadmap starting with internal platforms for employee use:
• Leveraging organizational data for Q&A sessions.
• Promoting knowledge sharing initiatives.
• Providing templates for content creation.
• Ensuring robust data governance compliant with PDPA and GDPR regulations.

Future Steps

Bitkub aims to:

• Transition to enterprise-grade solutions for external sales.
• Explore AI applications in blockchain, digital footprints, supply chains, and healthcare.

In conclusion, organizational leaders must grasp AI’s potential and leverage it for augmentation or automation to effectively transform their organizations. Merely directing without understanding is no longer sufficient.

Panel: Building and using AI models responsibly

Dr. Tiranee Achalakul from Big Data Institute (BDI) and Kobkrit Viriyayudhakorn from AIEAT/OpenThaiGPT provided insights into various aspects of leveraging AI responsibly and effectively.

Dr. Kobkrit discussed how AI, particularly GPT models like Copilot, enhances understanding of code, shifting roles from mere function and syntax knowledge to broader business comprehension. This evolution allows developers and data analysts to transition from creators to evaluators, emphasizing the need for solid fundamentals to ensure comprehensive coverage.

Dr. Tiranee highlighted the dual-edged considerations for AI users and creators. She emphasized the importance of careful data selection to avoid biases and ensure relevance, especially in legal contexts. Additionally, she underscored the necessity for rigorous verification of AI-generated content and the critical role of prepared, well-labeled datasets in governmental contexts.

Regarding the impact of AI on tech professions, Dr. Kobkrit cautioned against viewing AI as a replacement for human capabilities, stressing instead its role as a powerful tool for acceleration and enhancement. He discussed the risks and benefits of centralizing versus decentralizing AI deployment, advocating for skill development alongside AI integration to maximize its utility effectively.

Dr. Tiranee further elaborated on the necessity for enhanced technical and prompt skills to derive optimal benefits from AI tools. She emphasized the need for robust fundamentals and careful dataset curation to refine AI outputs and responses effectively.

In conclusion, Dr. Kobkrit recommended OpenThaiGPT as a pivotal resource for Thai language processing, developed collaboratively and leveraging substantial datasets for comprehensive training, including support from entities like Pantip and NECTEC.

This panel discussion highlighted the evolving landscape of AI utilization, emphasizing responsible practices and strategic skill development for maximizing its potential across various sectors.

In conclusion,

Image from K. Worranit Pisutakarathada

I would say… this would be my new experience of attending this tech-wellbeing meetup event since I only join on conference or tech demonstration like in the previous events. However, in this event, we could know many insights from many tech companies on their perspectives of using AI inside the organizations, which would be beneficial for us for organizing our own personal life until enterprise level.

Another thing that I couldn’t forget to mention would be about ‘getting swags’. One example that I would like to show to you guys would be getting free NFC card from PingCAP TiDB Cloud. I do have my own NFC card but it’s all plain white, so I decide to get an unique free one. Also, there would be one more booth from JetBrains as they give a free trial license on their software, I do register it to try on their software since I’ve used Visual Studio Code for a long time. I might give a review on its IDE later on.

Finally, P’Phumrapee and I finally advertised CreatorsGarten group for the final session of the event, as promoting on our upcoming event, which is held on the middle of October. Stay tuned for more information in CreatorsGarten page: https://web.facebook.com/creatorsgarten

And that’s the wrap of my story on Code Mania 1011 In Nutshell. In the next journey, we would go deeper down for other events that will show how powerful AI is, and how we can corperate with them in many ways from many experts, from many enterprise. Stay tuned for new stories and thank you for supporting me on writing articles for a long time.

Stay safe and have a good day.

Related blogs

• สรุปงาน Code Mania 1011: AI for Developers | naiwaen@DebuggingSoft from K. Chatri Ngambenchawong