AI/ML Engineer & Data Scientist

Hi,I'm

Muhammad Waleed

AIArchitectwith6+yearsofexperiencebuildingnext-generationAIsystems.SpecializinginLLMs,RAGsystems,ComputerVision,DataScience,andenterprise-gradeMLOpspipelines.

Muhammad Waleed - AI/ML Engineer
Scroll
Projects

Featured Projects

A selection of projects I've worked on, showcasing my skills across different technologies and domains.

Project 01

Predictive Demand Forecasting

Built predictive models using XGBoost for energy demand forecasting achieving 91% accuracy, preventing 200+ potential outages. Analyzed 5M+ consumption records with advanced SQL.

XGBoostForecastingSQLTime SeriesEnergy AI
View on GitHub
Predictive Demand Forecasting
Project 02

Healthcare Multi-Agent RAG System

Multi-agent RAG for $2B healthcare enterprise using LangGraph with custom ReAct reasoning and hybrid vector-graph knowledge base, processing 10k+ medical documents with FDA-compliant explainability.

LangGraphRAGHealthcare AIReActHIPAA
View on GitHub
Healthcare Multi-Agent RAG System
Project 03

AutoML Platform

Proprietary AutoML platform for SaaS startup implementing Neural Architecture Search with Bayesian optimization, automatically generating and deploying 200+ production models, reducing ML costs by 85%.

AutoMLNASBayesian OptimizationKubeflowMLflow
View on GitHub
AutoML Platform
Project 04

Production NLP Pipeline

Designed production NLP pipelines using spaCy and Hugging Face Transformers, processing 100K+ documents daily for multi-class classification with 60% latency reduction via ONNX optimization.

spaCyTransformersONNXNLPClassification
View on GitHub
Production NLP Pipeline
Project 05

Multi-Modal Foundation Model

Built multi-modal model combining Vision Transformer (ViT) with custom cross-attention layers for image-text understanding, achieving SOTA on 4 benchmarks. Published in NeurIPS 2024 workshop.

ViTCross-AttentionMulti-ModalPyTorchCLIP
View on GitHub
Multi-Modal Foundation Model
Project 06

Semantic Search System

Implemented semantic search using Sentence Transformers and FAISS, enabling sub-second retrieval across 1000+ documents with 88% MRR@10 for enterprise knowledge management.

Sentence TransformersFAISSSemantic SearchEmbeddings
View on GitHub
Semantic Search System
Project 07

AI Design Platform Pipeline

Custom diffusion pipeline for AI-powered design platform serving 50K+ users, implementing ControlNet, IP-Adapter, and LoRA fine-tuning, generating 50k+ images daily with 3-second latency.

ControlNetIP-AdapterLoRAStable DiffusionFastAPI
View on GitHub
AI Design Platform Pipeline
Project 08

Mixture-of-Experts Architecture

Pioneered novel MoE architecture with adaptive routing mechanism and constitutional AI alignment through custom RLHF pipeline, reducing inference costs by 67% while maintaining GPT-4 level performance.

MoERLHFPyTorchConstitutional AIvLLM
View on GitHub
Mixture-of-Experts Architecture
Project 09

Statistical Analysis Web Suite

Interactive applications for correlation, covariance, and descriptive statistics with real-time visualizations, export capabilities, and automated statistical testing.

StreamlitStatisticsData VisualizationPandasNumPy
View on GitHub
Statistical Analysis Web Suite
Project 10

Real-Time Voice AI System

Voice AI for US-based call center (5+ agents) using custom Whisper fine-tuning, streaming ASR with WebRTC, and multi-lingual NER, processing 100k+ hours monthly with 97% satisfaction.

WhisperWebRTCASRNERReal-time AI
View on GitHub
Real-Time Voice AI System
Project 11

INT4 Quantization Framework

Proprietary quantization technique combining INT4 weight quantization with knowledge distillation, enabling deployment of 70B models on consumer GPUs with only 2.3% accuracy degradation.

QuantizationINT4Knowledge DistillationONNXTensorRT
View on GitHub
INT4 Quantization Framework
Project 12

Enterprise RAG Microservice

Production-ready RAG microservice built with NestJS for ingesting PDF, DOCX, and TXT documents with semantic search powered by Milvus vector database and sub-second query latency.

NestJSMilvusRAGTypeScriptMicroservices
View on GitHub
Enterprise RAG Microservice
Project 13

n8n AI Workflow Automation

Enterprise AI automation workflows integrating LLMs, document processing, and multi-step reasoning pipelines. Scaled to 35+ enterprise clients globally with 100% on-time delivery.

n8nAI AutomationLLM IntegrationWorkflowsAPI
View on GitHub
n8n AI Workflow Automation
Project 14

Custom Diffusion Model Architecture

Engineered diffusion model with novel U-Net variants and CLIP-guided latent optimization, surpassing Stable Diffusion XL in FID scores while requiring 40% fewer parameters.

DiffusionU-NetCLIPComfyUIControlNet
View on GitHub
Custom Diffusion Model Architecture
Project 15

ETL Pipeline Automation

Automated ETL pipelines using Apache Airflow and Python, reducing manual reporting time from 20 hours to 30 minutes weekly with real-time data processing and monitoring.

Apache AirflowETLData EngineeringPythonAutomation
View on GitHub
ETL Pipeline Automation
Project 16

Churn Prediction Model

Implemented churn prediction using Random Forest with 92% precision, enabling proactive retention strategies. Analyzed 50K+ customer interactions using NLP techniques.

Random ForestChurn AnalysisNLPCustomer Analytics
View on GitHub
Churn Prediction Model
Project 17

Fortune 500 Custom 13B LLM

Architected and deployed custom 13B parameter LLM for Fortune 500 financial client using distributed DeepSpeed ZeRO-3, reducing document processing time by 94% with 94% accuracy.

DeepSpeedLLMFinancial AIZeRO-3Production ML
View on GitHub
Fortune 500 Custom 13B LLM
Project 18

Manufacturing Defect Detection

End-to-end computer vision pipeline for European manufacturing client combining custom YOLOv9 with Vision Transformer, achieving 90.97% accuracy across 50+ daily inspections on edge devices.

YOLOv9Vision TransformerEdge AIOpenCVTensorRT
View on GitHub
Manufacturing Defect Detection
Project 19

Patient Risk Prediction System

Developed patient risk prediction models using ensemble methods, improving early diagnosis accuracy by 35% across 10K+ patient records with real-time analytics dashboard.

Ensemble MethodsHealthcare MLPlotly DashRisk Prediction
View on GitHub
Patient Risk Prediction System
Project 20

100B+ Distributed Training Infrastructure

Designed distributed training system for 100B+ parameter models using Megatron-LM and DeepSpeed ZeRO-3, with custom CUDA kernels achieving 3.2x memory optimization across 128 A100 GPUs.

DeepSpeedMegatron-LMCUDAA100ZeRO-3
View on GitHub
100B+ Distributed Training Infrastructure
Project 21

Executive BI Dashboard Suite

Developed 15+ Power BI dashboards with DAX measures and real-time refresh, enabling C-suite executives to track KPIs and make data-driven decisions 40% faster.

Power BIDAXBusiness IntelligenceKPI Tracking
View on GitHub
Executive BI Dashboard Suite
Project 22

Neural Architecture Search System

End-to-end NAS system discovering optimal transformer variants, resulting in 3 patent applications. Architecture adopted by Fortune 500 client for production deployment.

NASAutoMLOptunaRay TuneBayesian Optimization
View on GitHub
Neural Architecture Search System
Project 23

Federated Learning Infrastructure

Federated learning system for global retail chain across 500+ locations, training models on-premise with differential privacy (ε=0.1), improving demand forecasting by 40%.

Federated LearningDifferential PrivacyEdge MLPyTorch
View on GitHub
Federated Learning Infrastructure
Project 24

Custom 7B Parameter LLM

Architected and trained custom 7B parameter LLM from scratch using RoPE and FlashAttention-2, achieving 15% better perplexity than LLaMA-2 with custom BPE tokenizer for multilingual code-switching.

PyTorchFlashAttention-2RoPECUDATransformers
View on GitHub
Custom 7B Parameter LLM