China · GMT+8

JamSylph

Algorithm Engineer | Multimodal AI Specialist

Specializing in computer vision and NLP, crafting intelligent systems that understand multimodal data

Learn More Contact Me

About Me

With an economics background and four years of experience in algorithm engineering, I've expanded my expertise from data analysis to computer vision and multimodal systems. Starting as a market researcher and data analyst in 2021, I quickly pivoted to computer vision algorithms and have been developing innovative AI solutions for real-world applications. My unique background combines data-driven insights with technical implementation, allowing me to develop algorithms that deliver both technical excellence and business value.

My Journey

After graduating with an economics degree in 2021, I began my career in market research and data analysis. Fascinated by the potential of AI, I transitioned to algorithm engineering with a focus on computer vision. My non-traditional background gave me a unique perspective on applying technical solutions to business problems. Over the years, I've developed expertise in building production-ready computer vision systems, and recently expanded into NLP and multimodal AI applications, implementing RAG systems with cutting-edge models like DeepSeek's R1.

Algorithm
Data
Innovation
Research

As an algorithm engineer, I develop and implement advanced machine learning models to solve complex problems. My expertise includes deep neural networks, reinforcement learning, and optimization algorithms. I specialize in transforming research concepts into efficient, production-ready code that can handle real-world challenges at scale. My work in computer vision and AI architectures has led to systems that achieve state-of-the-art performance while remaining computationally efficient.

With a strong foundation in data science, I transform raw data into valuable insights and actionable intelligence. My approach combines statistical rigor with practical business understanding, allowing me to identify patterns that matter. I've developed expertise in data preprocessing pipelines, feature engineering for machine learning, and building predictive models that drive decision-making. I'm particularly skilled in Retrieval-Augmented Generation (RAG) techniques and designing data architectures that enable AI systems to access and utilize knowledge effectively.

I thrive on pushing the boundaries of what's possible and finding novel solutions to challenging problems. My approach to innovation is both systematic and creative - I start by deeply understanding the problem space, then explore unconventional approaches that might yield breakthroughs. I believe in rapid prototyping and iterative development to test ideas quickly and refine them based on real feedback. Some of my most successful projects have come from combining techniques across different domains or applying established methods in new contexts. I'm particularly interested in innovations that make AI more transparent, efficient, and accessible.

I maintain a strong connection to industry innovations while focusing on practical applications. I regularly follow the latest developments in machine learning and contribute to projects through robust implementations and open-source collaborations. My technical interests include algorithm optimization, deployment efficiency, and creating systems that are both powerful and maintainable. I believe that combining industry best practices with cutting-edge approaches is essential for developing truly valuable solutions. I enjoy the challenge of transforming complex technical concepts into practical tools that create real business impact.

My Philosophy

I believe that technology should serve humanity by making our lives better, more productive, and more meaningful. My work is guided by a commitment to developing AI that is ethical, transparent, and accessible. I'm particularly interested in applications that have positive social impact, whether by making expert knowledge more available, automating tedious tasks, or providing new insights that weren't previously possible.

30+ Technical Solutions
20+ Completed Projects
4+ Years Experience

Core Strengths

Computer Vision Expertise

Four years of professional experience developing and optimizing computer vision algorithms for real-world applications.

NLP & RAG Implementation

Specialized in Retrieval-Augmented Generation using DeepSeek models to create context-aware AI systems.

Predictive Modeling

Background in economic forecasting and GM(1,1) gray prediction models, bringing quantitative analysis to AI solutions.

Multimodal Integration

Skilled at combining computer vision and NLP techniques to create comprehensive AI systems that understand diverse data types.

Portfolio Showcase

Featured Work

A selection of my most impactful projects and contributions across computer vision, NLP, and multimodal AI systems.

Computer Vision

Industrial Defect Detection System

Led the development of a high-precision computer vision system for detecting manufacturing defects, implemented across multiple production lines to enhance quality control processes.

99.2% Detection Accuracy
2023 Deployment Year
View Case Study
Computer Vision

Aerial Solar Panel Inspection System

Developed an autonomous drone-based inspection system for detecting defects in solar panel arrays using computer vision, employing YOLO object detection optimized for edge computing on drone hardware.

96.5% Detection Accuracy
80% Time Reduction
View Project
Computer Vision

Autonomous Vehicle Perception System

Developed a multi-modal perception system that fuses data from cameras, LiDAR, and radar to create a comprehensive understanding of the vehicle's surroundings in real-time.

98.7% Detection Accuracy
12ms Processing Time
View Case Study
NLP & LLMs

DeepSeek R1 RAG Implementation

Developed a domain-specific RAG system using DeepSeek's R1 model that integrates specialized knowledge bases to enhance context awareness and reduce hallucinations in generative AI responses.

45% Factual Improvement
2024 Implementation Year
Explore Project
NLP & LLMs / Research

Llama-3-7B Model Fine-tuning

Successfully reproduced and fine-tuned Meta's Llama-3-7B model for specialized industrial applications using parameter-efficient techniques. Created a custom evaluation framework for measuring hallucination rates and factual accuracy in industrial contexts.

37% Domain Performance Gain
4-bit Quantization Level
View Research
Multimodal AI

Vision-Language Manufacturing Assistant

Designed an integrated multimodal system that combines visual inspection with language processing to provide real-time assistance for manufacturing operations, enhancing both efficiency and accuracy.

32% Efficiency Increase
2024 Implementation Year
See Demo
Multimodal AI

Multimodal Sensor Fusion System

Pioneered a multimodal approach to quality inspection by fusing data from visual, thermal, and acoustic sensors. Developed custom neural architectures to integrate heterogeneous inputs for detecting complex defects not visible through a single modality.

42% Internal Defect Detection
28% Testing Reduction
View Project
Economic Analysis / Research

GM(1,1) Prediction Model

Academic research applying the GM(1,1) gray prediction model to accurately forecast manufacturer income and production metrics under uncertain market conditions, demonstrating the value of AI techniques in economic forecasting.

91% Prediction Accuracy
2021 Publication Year
Read Research
View Full Portfolio

Skills & Technologies

Computer Vision

OpenCV
PyTorch Vision
Object Detection
Image Segmentation
Visual Inspection

Natural Language Processing

DeepSeek Models
RAG Systems
Knowledge Graphs
Vector Databases
Semantic Search

Multimodal Integration

Vision-Language
Multimodal Fusion
Cross-Modal Learning
Feature Extraction
Transfer Learning

Tools & Economics

GM(1,1) Models
Manufacturing Analysis
Docker/K8s
Git/GitHub
MLOps

Projects

DeepSeek RAG Knowledge Assistant

An enterprise knowledge assistant built using DeepSeek R1 model with RAG (Retrieval-Augmented Generation) architecture for accurate and context-aware responses to company-specific queries.
Python DeepSeek R1 RAG
View Project 
# RAG Implementation
from deepseek import DeepSeekModel
from vector_store import VectorDB

def query_knowledge_base(query, context):
    # Retrieve relevant documents
    docs = vector_db.similarity_search(query)
    
    # Generate response with context
    response = model.generate(
        prompt=query,
        context=docs + context
    )
    
    return {
        "answer": response,
        "sources": [doc.metadata for doc in docs]
    }

Manufacturing Defect Detection

A real-time computer vision system that detects manufacturing defects with 98.7% accuracy. Implemented custom CNN architectures and data augmentation techniques to handle limited training data.
Python PyTorch Computer Vision
View Project 
class DefectDetector(nn.Module):
    def __init__(self):
        super(DefectDetector, self).__init__()
        self.features = nn.Sequential(
            nn.Conv2d(3, 64, kernel_size=3),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2),
            nn.Conv2d(64, 128, kernel_size=3),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2),
        )
        self.classifier = nn.Sequential(
            nn.Linear(128 * 54 * 54, 512),
            nn.ReLU(inplace=True),
            nn.Dropout(p=0.5),
            nn.Linear(512, 2),
        )

Economic Forecasting with GM(1,1)

Implementation of the GM(1,1) gray prediction model for manufacturing income forecasting. The model achieves a mean absolute percentage error of less than 3% on historical data.
Python Time Series Economic Modeling
View Project 
def gm11_forecast(x0, n):
    """
    GM(1,1) forecasting model
    x0: Original sequence
    n: Number of points to predict
    """
    x1 = np.cumsum(x0)
    z1 = 0.5 * (x1[:-1] + x1[1:])
    
    B = np.vstack([-z1, np.ones_like(z1)]).T
    Y = x0[1:].reshape((-1, 1))
    
    # Solve for a, b using least squares
    [[a], [b]] = np.linalg.inv(B.T @ B) @ B.T @ Y
    
    # Forecast future values
    x_pred = [(x0[0] - b/a) * np.exp(-a * (i)) + b/a 
              for i in range(len(x0) + n)]
    
    return x_pred
See More Projects

Contact

Interested in my work or have collaboration opportunities? Feel free to contact me!