SA
SAMEER ALIAI // SYSTEMS ENGINEER
Hire Sameer
SYSTEM READY•Open for AI Engineering Roles Worldwide
// APPLIED MACHINE LEARNING & MULTIMODAL SYSTEMS

Sameer Ali
Python AI Engineer

Multimodal Video Pipelines

Designing production-ready machine learning pipelines, fine-tuning parameter-efficient Vision-Language Models (4-bit QLoRA), and engineering asynchronous Python & FastAPI backends for real-world deployment.

Explore 8 Flagship SystemsLaunch AI Sandbox
8 Flagship
Core AI Projects
Production Built
10x Realtime
Pipeline Throughput
faster-whisper
4-Bit QLoRA
VLM Quantization
78% VRAM Saved
Sukkur IBA
CS Foundation
Graduating '26
PyTorchDeep Learning
faster-whisperSpeech AI
OpenCVComputer Vision
FastAPIBackend & APIs
DockerContainers
Qwen2-VL (QLoRA)Vision-Language
TensorFlowNeural Networks
spaCy NLPIntent & NLU
Next.js 14Modern Web
Python 3.11Core Language
PostgreSQLDatabase
PyTorchDeep Learning
faster-whisperSpeech AI
OpenCVComputer Vision
FastAPIBackend & APIs
DockerContainers
Qwen2-VL (QLoRA)Vision-Language
TensorFlowNeural Networks
spaCy NLPIntent & NLU
Next.js 14Modern Web
Python 3.11Core Language
PostgreSQLDatabase
8 FLAGSHIP PRODUCTION PROJECTS

Engineered AI & Software Systems

End-to-end architectures covering 4-bit Vision-Language Models, asynchronous multimodal video processing, autonomous agent search, and clinical deep learning diagnostics.

Autonomous & Agentic AI

Aegis-SWE: Autonomous Software Engineering Agent

Agentic MCTS Search, Tree-sitter Code RAG & Closed-Loop Docker Verification

An autonomous software engineering agent powered by Claude 3.5 Sonnet, GPT-4o, and Llama 3.3 engineered to resolve complex GitHub repository issues through guided search, AST code retrieval, and verified test execution.

Agentic MCTS Search: Employed Monte Carlo Tree Search with UCT (Upper Confidence Bound) to navigate code hypothesis branches with automated backtracking.
Code RAG: Leveraged Tree-sitter AST symbol call-graphs to retrieve relevant functions and symbols without overflowing context limits.
Prompt Engineering: Formulated a robust 3-role prompt system (Planner, Patch Generator, Critic) enforcing structured JSON diff formats.
Closed-Loop Verification: Executed patches inside Docker sandboxes to verify reproduction tests turned green before committing fixes.
PythonClaude 3.5 SonnetGPT-4oLlama 3.3MCTS / UCT SearchTree-sitter ASTDocker SandboxingJSON DiffsSWE-bench
Multimodal & Vision

Multimodal Video AI Pipeline

Speech Transcription, Semantic Topic Segmentation & Keyframe Extraction

High-performance asynchronous Python pipeline that transforms long-form unstructured video into structured, searchable chaptered formats with synchronized visual keyframes.

Designed a Python pipeline that converts long-form videos into structured chapters using speech transcription and semantic topic segmentation.
Implemented computer-vision keyframe extraction and concurrent processing to improve efficiency when handling long videos.
Debugged and integrated NLP and computer-vision components while maintaining modular, testable code and minimizing processing bottlenecks.
Pythonfaster-whisperOpenCVAsyncio / MultithreadingspaCy NLPNumPy
Deep Learning

Deep Learning Classification & Medical AI

Multi-Label Chest X-ray Classification & U-Net L*a*b* Image Colorization

Applied deep learning project training CNN classifiers on the NIH Chest X-ray14 dataset, alongside U-Net generative architectures for photo-realistic image colorization in L*a*b* color space.

Trained and evaluated multi-label classifiers on the NIH Chest X-ray14 dataset using MobileNetV2, ResNet50, and EfficientNetB0 architectures.
Implemented U-Net models for image colorization in the L*a*b* color space with perceptual loss metrics.
Benchmarked model performance using evaluation metrics (ROC-AUC, F1) and analyzed errors to identify opportunities for model improvement and bias reduction.
TensorFlowKerasPythonMobileNetV2ResNet50EfficientNetB0U-NetCNNs
Full-Stack AI

WebSense-AI Voice Control

Conversational Browser Extension with Intent Recognition & Low-Latency API

An intelligent voice-controlled web interaction platform utilizing NLP intent recognition and confidence scoring to navigate and execute browser actions through speech.

Collaborated on a conversational browser-extension project using natural-language understanding, intent recognition, entity extraction, and confidence scoring.
Built asynchronous Flask API endpoints to handle voice-driven requests and support low-latency communication between the AI backend and user interface.
Integrated client-side React frontend with real-time feedback audio status.
ReactNode.jsFlaskspaCy NLPPythonREST APIsEntity Extraction
Full-Stack AI

NexusStore — Modern E-Commerce & GenAI Concierge

Live Multi-Currency Sync (USD/PKR/EUR/GBP) & Groq LLM Shopping Assistant

Full-featured modern e-commerce web platform featuring real-time multi-currency & country synchronization and an intelligent Generative AI Shopping Concierge powered by Groq LLMs (120B/27B) with dynamic product card rendering.

Built a dual real-time multi-currency engine supporting USD ($), PKR (Rs), EUR (€), and GBP (£) with persistent locale switching.
Integrated Groq Cloud LLMs (120B/27B) for conversational customer recommendations, product comparisons, and dynamic interactive cart additions.
Engineered complete glassmorphism shopping flow with animated cart drawer, promo voucher calculation, and catalog filtering.
JavaScriptGroq GenAIMeta LLaMA 120BMulti-Currency EngineCSS GlassmorphismREST APIs
Multimodal & Vision

DocVLM: 4-Bit Fine-Tuned Vision-Language Model

QLoRA Parameter-Efficient Fine-Tuning on Qwen2-VL-2B for Structured Document Extraction

End-to-end multimodal deep learning pipeline that fine-tunes Qwen2-VL-2B via QLoRA (4-bit NF4) for zero-hallucination structured JSON extraction from complex invoices, tabular receipts, and forms.

Fine-tuned Qwen2-VL-2B cross-attention projection layers with PEFT QLoRA (r=16, alpha=32) while freezing 4-bit base weights (<0.8% trainable parameters).
Achieved 96.8% JSON schema compliance and 94.2% key-value extraction F1 on unseen document test sets.
Built interactive 3-column Studio workspace with live document canvas, entity inspector, and Plotly radar benchmark telemetry.
PyTorchPEFT QLoRABitsAndBytes (4-Bit NF4)Qwen2-VL-2BTransformersStreamlitPlotly
Deep Learning

Hematological Disease Diagnostic System (Anemia & Leukemia)

Dual-Disease Clinical Diagnostic Classifier using XGBoost & LightGBM with 100% Recall

Dual-disease clinical diagnostic pipeline utilizing Gradient Boosting Machines (XGBoost & LightGBM) to detect Anemia and Leukemia from routine Complete Blood Count (CBC) markers and Bone Marrow Blast percentages with zero false negatives.

Engineered dual classification pipelines with XGBoost (Leukemia) and LightGBM (Anemia) utilizing clinical blood markers (Hemoglobin, MCV, MCH, MCHC, and Blast counts).
Resolved accuracy paradox in medical screening through target realignment, SMOTE resampling, and recall-weighted cost-sensitive learning to achieve 100% leukemia recall.
Generated clinical explainability artifacts including feature importance rankings, confusion matrices, and correlation heatmaps to support clinician decision-making.
PythonXGBoostLightGBMScikit-LearnSMOTEPandasNumPyMatplotlib / Seaborn
Multimodal & Vision

AI Image Enhancement & Colorization Studio

OpenCV DNN Colorization (CIE LAB) & MIRNet-Inspired Low-Light Enhancement

Dual-pipeline visual restoration web application combining OpenCV DNN (Zhang et al. ECCV 2016 Caffe) for photo-realistic CIE LAB grayscale colorization and MIRNet-inspired low-light contrast & gamma enhancement with a unified Flask inference server.

Implemented automatic grayscale colorization by extracting CIE LAB L-channel and predicting ab chroma channels via pre-trained Caffe models in OpenCV DNN.
Engineered a lightweight MIRNet-inspired low-light image enhancer combining CLAHE, noise filtering, and gamma correction for real-time CPU deployment.
Built and deployed a unified Flask inference API server and responsive web studio for interactive photo restoration.
PythonOpenCV DNNCIE LAB Color SpaceFlaskNumPyMIRNetPillow
INTERACTIVE EVALUATION PLAYGROUND

Live AI Pipeline & Model Sandbox

Simulate how Sameer executes multimodal video processing pipelines, evaluates medical deep learning classifiers, and extracts NLP intent entities.

sameer@eval-engine: ~/doc-vlm-qlora
sameer@eval:~$python -m doc_vlm.inference --weights qwen2-vl-2b-qlora-nf4.pt --doc sample_invoice.pdf --format json
[DOC-VLM] Loading 4-bit Quantized Vision-Language Model...
[VRAM FOOTPRINT] Base Weights: 1.42 GB | QLoRA Adapters: 0.40 GB | Total VRAM: 1.82 GB (78% VRAM Reduction)
[INFERENCE] Visual Encoder (ViT-H/14) processing document image at 1080p...
[DECODER] Cross-attention projection generating structured JSON tokens:
{
  "document_type": "Commercial Tax Invoice",
  "invoice_number": "INV-2026-0884",
  "vendor": "ACME Supplies & Logistics Ltd",
  "date": "2026-10-24",
  "line_items": [
    { "desc": "Cloud Compute Cluster H100 (x4)", "amount": 4200.00 },
    { "desc": "Neural Pipeline Storage (10TB)", "amount": 675.00 }
  ],
  "total_due": 4875.00,
  "currency": "USD",
  "model_confidence": 0.997,
  "schema_validation": "PASSED (100%)"
}
[PERFORMANCE] Inference Latency: 34.2ms | Tokens/sec: 48.5 tps
Status: Evaluation Criteria Active•Backend: Python / MCTS / Tree-sitter / Docker
Click tabs above to switch evaluation tasks ↗
TECHNICAL PROFICIENCY & STACK

Engineering & AI Capabilities

Specialized toolkit spanning LLM evaluation benchmarks, deep neural network training, asynchronous Python architectures, and production computer vision.

Agentic AI & LLM Systems

5 core competencies
Autonomous Coding AgentsExpert

MCTS search, UCT hypothesis trees, automated backtracking

Code RAG & AST Call-GraphsAdvanced

Tree-sitter AST syntax parsing, symbol dependency mapping

Multi-Agent Protocols & PromptingAdvanced

Planner-Generator-Critic roles, JSON diff generation

Docker Sandboxing & SWE-benchAdvanced

Closed-loop verification, reproduction test execution

Model OrchestrationAdvanced

Claude 3.5 Sonnet, GPT-4o, Llama 3.3, Groq API

AI, Deep Learning & Vision

5 core competencies
Deep Learning ArchitecturesAdvanced

CNNs, MobileNetV2, ResNet50, EfficientNet, U-Net

Computer Vision & Video ProcessingAdvanced

OpenCV, keyframe extraction, L*a*b* color spaces

Speech AI & Audio PipelinesAdvanced

faster-whisper, audio segmentation, transcription

NLP & Intent RecognitionAdvanced

spaCy token pipelines, entity extraction, confidence scoring

Prompt Engineering & GenAIProficient

Google Prompting Essentials, Chain-of-Thought, System Prompts

Backend, APIs & Cloud Systems

5 core competencies
FastAPIAdvanced

High-performance async REST APIs, Pydantic data validation

FlaskAdvanced

Lightweight microservices & AI inference endpoints

DockerAdvanced

Containerization of ML pipelines & reproducible environments

Asynchronous ProgrammingAdvanced

asyncio, multithreading, concurrent queues, background workers

REST APIs & WebSocketsAdvanced

Low-latency bidirectional AI communication

Programming & Software Engineering

5 core competencies
PythonExpert

Production pipelines, async, OOP, typing, NumPy

JavaScript / TypeScriptProficient

Modern React, Node.js, Next.js

Unit & Integration TestingAdvanced

pytest, test-driven validation, CI/CD

Debugging & Code RefactoringAdvanced

Profiling, memory optimization, modular design

Git & GitHubAdvanced

Branching strategies, collaborative workflows

Frameworks, Libraries & Tools

5 core competencies
TensorFlow & KerasAdvanced

CNN classifiers, U-Net models, transfer learning

OpenCVAdvanced

Keyframe extraction, frame diffing, image processing

spaCy NLPAdvanced

Custom entity extraction, rule-based matching

faster-whisperAdvanced

Accelerated speech-to-text inference

StreamlitAdvanced

Rapid AI demo prototyping and dashboarding

CREDENTIALS & CONTINUOUS LEARNING

Certifications & Industry Specializations

Verified coursework from Google, DeepLearning.AI, IBM, and top global institutions.

GoogleCompleted

Google Prompting Essentials

Skills Acquired:
Prompt DesignIn-context LearningEvaluation Principles
University of Michigan / CourseraCompleted

Python 3 Programming Specialization

Skills Acquired:
Advanced Data StructuresAPIs & Web ScrapingPillow & OCR
GoogleCompleted

Google AI Essentials

Skills Acquired:
GenAI FoundationsResponsible AI PracticesWorkflow Optimization
DeepLearning.AI / Andrew NgIn Progress

DeepLearning.AI Machine Learning

Skills Acquired:
Supervised & Unsupervised LearningNeural NetworksRegularization
IBMIn Progress

IBM AI Engineering

Skills Acquired:
Deep Learning & KerasPyTorch & TransformersEnterprise ML Pipelines
ACADEMIC FOUNDATION

Education & Core Foundations

Rigorous training in computer science, algorithmic design, and artificial intelligence at one of Pakistan's leading institutions.

Undergraduate Degree

Bachelor of Science in Computer Science

Sukkur IBA University

Graduating 2026Pakistan

Core focus on Algorithms, Artificial Intelligence, Data Structures, Operating Systems, Computer Vision, and Distributed Computing.

Key Focus Areas:
Artificial Intelligence & Machine LearningData Structures & AlgorithmsOperating Systems & MultithreadingDatabase Systems & Query OptimizationComputer Vision & Pattern RecognitionSoftware Engineering Principles
LET'S CONNECT & COLLABORATE

Get In Touch

Interested in discussing AI engineering roles, LLM evaluation pipelines, or multimodal systems? I'm open to remote full-time opportunities and impactful collaborations.

EMAIL ADDRESS
sameer2659110@gmail.com
PHONE / WHATSAPP
0300 3533809
LOCATION & TIMEZONE
Hyderabad, Pakistan

Open to Remote, Hybrid, and On-site opportunities (Flexible worldwide overlap & relocation).

GitHub LinkedIn

Send a Direct Message

Fill out the details below to trigger a pre-formatted email directly to Sameer.

Direct transmission powered by EmailJS • Also reach Sameer directly at