Professional experience across AI and game development
AI Engineer • Game Developer • Dhaka, Bangladesh
Nayeem Hossen Jim
I build AI systems and gameplay systems that are meant to ship.
AI Engineer and Game Developer with 1+ year of experience building and deploying RAG systems, fine-tuning LLMs, and engineering Unreal Engine 5 gameplay mechanics. Skilled in Python, C++, LangChain, LangGraph, and Hugging Face Transformers, with applied MLOps and LLMOps experience in production work.
Current focus
RAG, LLM fine-tuning, Unreal Engine 5, Unity
Proof points
Impact at a glance
Reduction in inference latency after LoRA and QLoRA fine-tuning
Accuracy for the real-time skin cancer detection system
About
Built for production outcomes
What I build
RAG applications, LLM fine-tuning workflows, computer vision systems, and gameplay systems that need to work in real use, not just in a demo.
How I work
I combine model development, deployment, and iteration with enough software discipline to ship production work without losing speed.
Profile snapshot
- SummaryAI Engineer and Game Developer with production RAG, fine-tuning, and Unreal Engine experience
- LocationDhaka, Bangladesh
- Primary stackPython, C++, LangChain, LangGraph, Hugging Face
Skills
Core strengths
Programming Languages
- Python
- C++
- C
- C#
AI & Machine Learning
- LangChain
- LangGraph
- Hugging Face Transformers
- TensorFlow
- PyTorch
- RAG Systems
- LoRA
- QLoRA
- Unsloth
- YOLO
MLOps & LLMOps
- MLflow
- DVC
- Docker
- CI/CD Pipelines
- Model Versioning
- Experiment Tracking
- Production Monitoring
- n8n
Game Development
- Unreal Engine 5
- C++ Gameplay Systems
- Blueprints
- Unity
- Blender
- ShaderLab
- HLSL
Databases & APIs
- PostgreSQL
- pgvector
- MySQL
- FastAPI
- REST APIs
Developer Tools
- Git
- GitHub
- VS Code
- Google Colab
- Kaggle
Featured Projects
Projects
01
Language modeling
LLM From Scratch
PyTorch • Python
Built a GPT-3-architecture language model from scratch in PyTorch and fine-tuned it for personal assistant tasks.
- Custom SimpleTokenizerV2 with regex-based cleaning, vocabulary mapping, and unknown token handling.
- Processed 50,000+ tokens per second on consumer hardware.
02
Efficient fine-tuning
Fine Tune LLM
Unsloth • Hugging Face • LoRA
Applied LoRA and QLoRA to fine-tune large language models with a lighter training footprint and faster iteration cycles.
- Reduced GPU memory usage by 60% compared with full fine-tuning baselines.
- Cut training time to less than 20 minutes for new fine-tuning jobs.
03
Computer vision
Skin Cancer Detection
YOLO • FastAPI • JavaScript
Developed a full-stack skin cancer detection system with model deployment, API integration, and a browser-based frontend.
- Reached 91% detection accuracy across 5 cancer categories in the test dataset.
- Deployed on a VPS with sub-200 ms inference for live use.
04
Multiplayer systems
Multiplayer Plugin
Unreal Engine 5 • C++
Built a modular multiplayer plugin in C++ for Unreal Engine with a focus on reusable architecture and multiplayer readiness.
- Designed for Steam and Xbox-style multiplayer integration paths.
- Structured to support reusable plugin-based workflows.
05
Gameplay systems
The Last Gods
Unreal Engine 5 • C++ • Blueprints
Developed a third-person action game in Unreal Engine 5 with dynamic combat, responsive controls, and immersive gameplay systems.
- State-machine-driven enemy boss AI with 8 distinct attack patterns.
- Reduced player loop repetition by 50% in playtesting sessions.
Experience
Current role and impact
October 2025 - Present
Junior AI Engineer | Softvence Delta, Dhaka, Bangladesh
- Architected and deployed production-grade RAG systems using LangChain and LangGraph, processing 100,000+ queries per month and driving a 30% increase in automated customer issue resolution.
- Fine-tuned Hugging Face Transformer models with LoRA and QLoRA for domain-specific NLP tasks, achieving a 40% reduction in inference latency and measurable gains in task accuracy.
- Established MLOps best practices across MLflow, DVC, CI/CD pipelines, and production monitoring, cutting deployment errors by 25%.
- Built AI automation workflows with n8n and 10+ third-party API integrations to remove manual data-handling steps.
- Developed a Bangla Sign Language detection model at the alphabet level using a custom YOLO dataset, reaching 87% detection accuracy in live-camera tests.
Education & Certifications
Education and credentials
Education
- B.Sc. in Computer Science and Engineering, Uttara University, Dhaka, Bangladesh
- CGPA: 3.5 / 4.0
Certifications
- AWS AI Practitioner Challenge – Udacity
- Advanced Learning Algorithms – Stanford University (Coursera)
- Supervised Machine Learning: Regression and Classification – Stanford University (Coursera)
Awards & Leadership
- 12th Place – Game Jam Competition, BUET CSE Fest 2024
- Former Vice President – Machine Learning Club, Uttara University