Green Software Engineering and Digital Sustainability

Personal Tech Sandbox & Green Architecture

As my personal technical sandbox designed for high-availability and secure enterprise scenarios, this demo explores how to build highly efficient, robust, and low-energy enterprise digital systems with a minimal carbon footprint. I embed AI-native solutions deep into core architectures, aligning with UN SDGs 9 and 12, to provide actionable architectural examples of green software engineering.

Digital Carbon Calculator Explore AI Suite

AI Innovation & Security

AI-Native & Saddle Governance

Combines Text-to-SQL logic with the llmSqlGuard safety layer to place strict business constraints on LLMs. Automatically intercepts SQL injection and unauthorized data access at the application layer.

Orchestration Layer

Tower Crane Agentic Flow

Implements the self-elevating Tower Crane architecture (Plan-Act-Log-Refine) allowing autonomous agents to plan, execute, and verify code within strict test contracts under human governance.

Green Engineering & Governance

Green DevOps & ADO

Minimizes digital carbon emissions through payload optimizations, while integrating dependency audits and regression tests directly into Azure DevOps pipelines via self-hosted build agents.

Enterprise Exclusive Proposal AI-Native Semantic Retrieval Proof of Concept (PoC)

Enterprise AI-Native Semantic Retrieval Sandbox

Next-Gen Intelligent Semantic Retrieval System based on Four Pillars (Text-to-SQL + Security Guard)

Tailored for enterprise-scale and security compliance retrieval requirements, demonstrating how to break through AI landing bottlenecks via "Real-time Feature Pipelines (Lakehouse)" and "Model Constraints (Saddle Governance)":

1. MADAGA (Semantic Search Text-to-SQL)

Seamlessly connects to Lakehouse, translating natural language search queries into SQL. Integrates local in-memory TF-IDF sparse retrieval (Schema Linking) into Context Engineering, reducing schema mapping latency to <1ms to guarantee rapid online responses and align with sustainable, low-entropy metrics.

2. LLM-SQL-Guard (Security Saddle for PII Protection)

Performs AST grammar validation before sending generated SQL to the database. Features built-in PII Gate anonymization technology, physically blocking unauthorized queries (e.g., non-privileged accounts accessing sensitive fields) and malicious injection, ensuring sensitive database entries remain securely within boundary.

3. Semantic Cache (High-Frequency Semantic Query Cache)

Intercepts high-frequency repetitive search intents (e.g., "Find Frontend", "Material Supplier") at the Gateway layer using vector similarity. Bypasses over 40% of LLM calls, achieving millisecond-level responses and lowering GPU workload.

Retrieval AI-Terminal

# Step 1: Text-to-SQL via MADAGA

$ madaga --query "Find React developers in Taipei with Docker experience"

Generated SQL:
SELECT name, skills FROM resumes WHERE location = 'Taipei' AND skills LIKE '%React%' AND skills LIKE '%Docker%';

# Step 2: AST & PII Safety via LLM-SQL-Guard

$ sqlguard --check "SELECT phone, address FROM resumes"

[VIOLATION] Access Denied: Role 'Recruiter' has no permission for PII data. Request Blocked!

# Step 3: Semantic Cache Lookup (Cost Optimization)

$ cache --query "Find React developers"

[HIT] Similarity: 98.2%. Bypassing LLM. Returning cached result.

BLOCKED CACHE HIT

Latency: 2ms | GPU Load: 0%

AI-Native & Digital Transformation: 4 Core Capabilities

🌐

1. Core System Architecture & Engineering

Lakehouse Real-time Data Pipelines: Breaks the batch latency of Offline DBs, constructing low-latency, highly consistent real-time feature engineering pipelines to support the high concurrency demands of next-gen intelligent matching.
Service Interfaces & Data Flow: Standardizes service API interfaces and structured data transmission pipelines to ensure efficient microservice collaboration.
Forward-looking Design & Tech R&D: Drafts comprehensive architectural designs and specs to resolve technical bottlenecks and establish coding standards.

⚡

2. AI Infrastructure & Compute Planning

AI Saddle Governance: Imposes business logic and safety constraints on high-entropy algorithms through security guardrails and semantic caching, ensuring stable and cost-effective model deployment in production environments.
Automated MLOps & LLMOps: Builds standardized pipelines from research and training to deployment and delivery, ensuring reliable production releases.
Model Deployment & Tooling: Provides foundational integration for open-source models and commercial APIs, optimizing inference speeds.
Lifecycle Model Monitoring: Works alongside algorithm teams to define evaluation metrics and real-time feedback loops for LLM quality.

🤝

3. Cross-Departmental & Technical Strategy

Search & Recommendation Integration: Collaborates with product and data science teams to perform feasibility assessments and engineering optimizations for high-traffic recommendation scenarios.
Algorithm Integration SOPs: Formulates standardized integration SOPs and setups real-time performance tracking for search models.
Next-Gen Tech Stacks: Guides engineering teams to expand their architectural horizons, driving the development of advanced enterprise AI/ML toolchains.

🚀

4. Project Execution & Technical Communication

Agile Execution of Key Projects: Deconstructs complex requirements into concrete milestones, ensuring high-quality and on-time engineering delivery.
Bridges Technical Disciplines: Achieves optimal balance among algorithm theories, hardware limits, and business goals to facilitate cross-functional alignment.
Intelligent Agents & Automation: Scales AI implementation, including building smart agents, automating complex business workflows, and streamlining tools.

Aligning with SDG 9 & SDG 12: Digital Asset Sustainability & Transformation

In today's fast-paced digital era, internet services and data centers generate over 3.7% of global greenhouse emissions. As software engineers, it is our responsibility to rethink software architecture—shifting from a purely feature-driven focus to a "green efficiency-first" mindset. Through AI-native technologies, we open up higher-efficiency pathways for corporate digital transformation.

AI-Native & Digital Transformation

We weave AI-native concepts directly into digital transformation. Rather than treating AI as an add-on, we architect intelligent systems around autonomous agents. While driving business agility, we emphasize:

Green Inference: Optimizing prompts and inference strategies to process targets with minimal tokens, slashing GPU/CPU power usage.
Smooth Refactoring: Utilizing low-energy microservices and containerization to help companies smoothly decommission monoliths and complete green transformations.

Low-Entropy Code Practices

We write code according to the "Principle of Minimal Entropy", meaning:

No Redundant Payloads: Striking strict API formatting contracts, preventing malformed encodings, and conserving bandwidth.
Algorithmic Complexity Control: Keeping core business calculations below O(n) complexity to avoid overloading server resources.
Lightweight Server-side Rendering: Relying on Thymeleaf instead of heavy client-side JavaScript frameworks, drastically lowering customer browser rendering energy.

Software Integrity Framework

True software sustainability stems from maintainability. We validate code revisions through automated pipelines, generating traceable evidence of security and performance compliance. This ensures enterprise systems remain healthy and adaptable for the next decade, fulfilling the resilient digital infrastructure goal of SDG 9.

🌱 Live Digital Carbon Calculator

This page has been optimized for low-carbon delivery (including converting images to high-compression WebP, lazy loading resources, and removing redundant scripts). Here are the live sustainability metrics for this visit:

0.000

Estimated carbon saved per load (g CO₂e)

0.0

Optimized page size (KB)

Aligned with SDG 12 (Responsible Consumption) Based on the Sustainable Web Design model

✉ Contact Us

If you are interested in my AI-native solutions, LLM-SQL-Guard architecture, or general technical discussions, please leave your details. I will reach out to you as soon as possible.

Name *

Email *

Subject *

Message *

About Us

Dedication to sustainable enterprise software and artificial intelligence:

Mark Wong

25+ years of enterprise software engineering
Education:
- M.S. in Computer Science & Information Engineering, Tamkang University (2023)
  Thesis: Text-to-SQL Query Generation using Deep Learning (Written in Chinese)
- B.S. in Civil Engineering, National Cheng Kung University (1996)
Certificates: SCJP, SCWCD Certified Developer
Training: Project Management Professional (PMP) Program, Institute for Information Industry
Journal Articles: Information Retrieval & Full-Text Search Practice (NTU Buddhist Library) (PDF in Chinese)
Focus: Deep learning, Natural Language Processing, Full-Text/Vector Retrieval
Hobbies: Mountaineering, Cinema

R&D Achievements

From Low-Level Algorithms to Next-Gen AI Search

Demonstrating strong technical agility and architectural transformation, addressing enterprise demands for Big Data & AI:

Full-Text Search Practice (NTU Buddhist Digital Library)

Led the system architecture upgrade by replacing database relational search with Lucene full-text search. This built a deep foundation in Inverted Index, tokenization, and query performance tuning, translating directly into vector retrieval optimizations for modern RAG architectures, recently expanded with Lucene 9, HNSW graph search, and Project Panama (SIMD) vector acceleration.
Read Journal Article (PDF in Chinese)

High-Precision Algorithm R&D (Navigation)

Served as Senior Engineer in Hsinchu Science Park, developing map compression algorithms for navigation devices. Under extreme embedded resource constraints, gained deep insight into system performance, memory management, and massive spatial data read optimization.

Smart Manufacturing & AI Model Landing (MES)

During tenure at GSS, validated and engineered the landing of AI prediction models for smart dyeing. Here, I deeply realized the Dual Bottleneck of transitioning from "post-hoc analysis" to "real-time prediction", forging my unique architectural philosophy of governing AI algorithms with "physical/business constraints (Saddles)" and "real-time streaming pipelines."
Read Media Coverage & Interview (MakerPRO) (in Chinese)

IoV & AIoT Cloud Infrastructure

Led AWS cloud architecture design and high-availability (HA) configuration, implementing massive vehicle data synchronization based on the MQTT protocol. Maintained ultra-high stability under high-concurrency, low-latency, and large-scale IoT data throughput scenarios.

Technical Portfolio

madaga
Enterprise-grade Text-to-SQL & RAG architecture using Spring Boot 3, integrating Lucene 9 + HNSW vector search, accelerated with Project Panama FFM & SIMD.
llm-sql-guard
AI application security defense layer (Text-to-SQL diagnostics and AST validation gate).
min-char-irnn (java-lstm)
Java implementation of long-sequence RNN/LSTM materials dynamics engine with rotational equivariance and physical constraints, accelerated via Vector API SIMD and Project Loom.