I am an AI Research Engineer with hands-on experience designing, building, and deploying scalable AI/ML systems, working across academic research, industry R&D, and cloud-based AI infrastructure.
My expertise spans Large Language Models (LLMs), Generative AI, Federated Learning, Multimodal Systems, Time-Series Deep Learning, and Backend + MLOps engineering.

I am passionate about solving large-scale AI problems by bridging research, software engineering, and real-world deployment.

South Korea | Federated Learning, Edge AI, Privacy-Preserving ML

I conduct advanced research in Federated Learning (FL) focused on AI training across resource-limited, geographically-distributed edge devices.
My work addresses real-world challenges such as communication constraints, non-IID data, and satellite/LEO-based learning scenarios.

• Designing communication-efficient FL algorithms to reduce uplink/downlink latency.
• Implementing privacy-preserving learning mechanisms, including:
  • Differential Privacy (DP)
  • Secure Multi‑Party Computation (SMPC)
  • Homomorphic encryption techniques
• Developing simulation frameworks using PyTorch and TensorFlow Federated.
• Studying robustness under client dropouts, device mobility, and non-IID distributions.

• LEO satellite‑based federated learning
• Split Learning + Federated Distillation
• Communication-efficient gradient compression
• Secure and scalable FL protocols

USA (Remote) | Multimodal AI Systems, Backend, MLOps

At Noctal, I led the architecture and development of advanced AI pipelines, scalable microservices, and production-ready backends for multimodal AI applications involving audio, vision, and text.

• Deployed state-of-the-art models:
  • Vision Transformers (ViT)
  • Audio Spectrogram Transformers (AST)
  • Stable Diffusion for text-to-image generation
• Built multimodal RAG systems combining:
  • Audio embeddings
  • Video frame understanding
  • Text retrieval + LLM reasoning
• Engineered vector search systems using FAISS, Pinecone, Weaviate.
• Designed full backend pipelines using:
  • FastAPI, Django, GCP (Vertex AI), Docker, Kubernetes
• Optimized large-scale inference on GKE Autopilot with GPU/TPU nodes.

• Leading AI/ML microservice development
• Architecting cloud-ready AI pipelines
• LLM fine-tuning and domain adaptation
• High-performance model inference engineering
• Building scalable backend + infrastructure for AI products

Islamabad (Remote) | Mobile AI, CV, Diffusion Models, AWS

I led the AI initiatives to build next-generation mobile AI applications, from research and prototyping to deployment.

• Developed end-to-end AI tools for:
  • Face swapping
  • Image enhancement
  • Edge detection using YOLO
  • Image scanning and classification
  • Text-to-image generation using diffusion models
• Built scalable cloud inference systems using AWS Lambda, AWS Kubernetes, CI/CD pipelines.
• Conducted research on quantization, enabling high-performance on-device LLMs.
• Reduced cloud API expenses by migrating to in-house ML models.
• Mentored developers for AI integration into mobile apps.

Islamabad | NLP, Computer Vision, RAG, Time-Series ML

Worked on multiple domain-specific AI applications, focusing on NLP, CV, and predictive analytics.

• NolixAI
  Edge-device AI (TensorFlow Lite, CNNs, Raspberry Pi) for leak detection and embedded automation.
• Oddson
  NLP + LLM-based RAG chatbot using:
  • FastAPI
  • Pinecone vector DB
  • MongoDB + Redis
  • AWS services
• Caroogle
  Vehicle data scraping + ML-based price prediction, risk analysis, and recommendation system.

Islamabad | Time-Series ML, Forecasting, Industrial AI

Focused on building AI systems for Industry 4.0, especially time-series forecasting and anomaly detection.

• DeepAR
• ARIMA
• XGBoost
• TCN
• Transformer-based models (TFT, Informer)
• RNN/LSTM-based forecasting

• Built full ML pipelines: data → modeling → deployment
• Collected and analyzed accelerometer data for anomaly detection
• Developed forecasting systems for manufacturing supply chains
• Created generative AI-based reporting pipelines
• Ensured robustness and optimization of deployed AI models

• Large Language Models (LLMs)
• NLP, Tokenization, Embeddings
• Generative AI (Diffusion Models, LLM Fine-Tuning)
• Vision Transformers, CNNs, AST
• Time-Series Forecasting (DeepAR, TCN, TFT)
• Federated Learning (FL), Split Learning, Knowledge Distillation
• Multimodal AI (audio + vision + text)

• FastAPI, Django, Flask
• Docker, Kubernetes, GKE Autopilot
• AWS (Lambda, S3, EC2, EKS)
• GCP (Vertex AI, GKE, Cloud Run)
• CI/CD, GitHub Actions

• Pinecone, Weaviate, FAISS
• MongoDB, MySQL, Redis
• Ray, Celery task queues
• Web scraping: Playwright, Selenium, Scrapy

• Python
• PyTorch, TensorFlow, Hugging Face
• ONNX Runtime, TensorRT
• RAG frameworks, prompt engineering

• LinkedIn: https://www.linkedin.com/in/iffishells/
• Portfolio: https://iffishells.wordpress.com/
• Email: iffiskhells@gmail.com