Umer Aftab

AI Engineer specializing in Agentic RAG, Generative AI, and Production-Ready ML Systems.

Based in: Lahore, PakistanMost recently: Machine Learning Engineer, ForrofExperience: 1 year

Umer Aftab is an AI Engineer with extensive experience in designing and deploying end-to-end machine learning systems, including multi-agent RAG architectures and medical imaging research. He possesses a robust technical stack spanning PyTorch, LangGraph, and Dockerized API deployment to deliver high-performance AI solutions.

Experience

Machine Learning Engineer

ForrofRemote, PakistanworkAug 2025 – Feb 2026

Summary

Delivered client-facing AI/ML solutions across computer vision and NLP, owning the full lifecycle from data preparation and model development through deployed, client-facing REST APIs. Built and optimized deep learning models in TensorFlow and PyTorch on cloud GPUs, tuning architectures and training pipelines for accuracy and inference speed. Shipped production-ready workflows with cross-functional teams using Docker-based deployment and REST API integration.

What I did

Owned the end-to-end lifecycle of a client's RAG-based document assistant, including data ingestion, retrieval pipeline, FastAPI service, and Dockerized deployment.

TensorFlowPyTorchDockerREST APICloud GPUsChromaDBFastAPI

Skills

Technical

Pandas

PyTorch

Python

RAG Pipelines

NumPy

Agentic AI

LangChain

LangGraph

Git/GitHub

TensorFlow

Scikit-learn

OpenCV

YOLOv5

Transformers

Hugging Face

Fine-tuning

ChromaDB

Prompt Engineering

Claude

BERT

spaCy

NER

Semantic Search

Text Classification

FastAPI

Docker

GPT

Streamlit

Matplotlib

Seaborn

Plotly

Celery

Redis

SSE Streaming

JWT Auth

PostgreSQL

Explainable AI (Grad-CAM)

Keras

MySQL

Selenium

Langfuse

Gemini

Mem0

TensorBoard

BeautifulSoup

Scrapy

Nginx

SQL

C++

Java

Projects

YouTube Knowledge Assistant

Developerpersonal

Built a multi-video RAG chatbot with timestamped source-grounded answers, cross-video comparison, comment sentiment analysis, and quiz/flashcard generation.

RAGLangChainChromaDBGPT-4o-mini

View the code

Deepfake Forgery Detection

Developeracademic

Designed a hybrid VGG16 + BiLSTM model detecting forged video frames with 94% frame-level accuracy on large-scale annotated video datasets.

VGG16BiLSTMVideo Analysis

View the code

RecruitIQ - AI Resume Screening

Developerpersonal

Built an end-to-end pipeline ranking candidates against job descriptions via semantic similarity, multi-layer skill matching, and BERT-based NER identity extraction.

NLPBERTspaCyNER

View the code

Ocular Disease Diagnosis with Diffusion-Based Augmentation

Research Leadacademic

Designed six experimental cases on ODIR-5K comparing diffusion-based augmentation strategies for disease classification, with Grad-CAM explainability analysis.

PyTorchVGG19Grad-CAMDiffusion Models

AI-Powered Smart Dustbin

Developeracademic

Built a smart waste system pairing real-time CV classification (∼91% accuracy) with IoT sensors to automate detection, segregation, and collection scheduling.

Computer VisionIoTYOLOv5

View the code

Product Recommendation Agent

Architect & Developerpersonal

Architected a complete production system: a 9-node LangGraph agent with parallel RAG retrieval, live price grounding (Tavily), and a custom confidence-scoring engine producing reasoning traces per recommendation. Deployed end to end on Docker + Nginx with Celery task queues, real-time Redis SSE streaming, Mem0 memory, JWT-authenticated APIs, and Langfuse observability.

LangGraphFastAPIClaudeChromaDBTavilyDockerNginxCeleryRedisSSE StreamingMem0JWTLangfuse

View the code