LAYER 00 · INPUT / machine learning engineer

MuhammadM.Sc. · CS SaadNajib

I build systems that see. From a published image-colorization network to face-recognition pipelines and 3D human understanding — I turn pixels into decisions.

Explore the work Download résumé
  • 0years in AI & software
  • 0vision projects shipped
  • 0published paper
Muhammad Saad Najib
FIG. 00 — hover / touch the portrait to see what the network sees
scroll

LAYER 01 · FEATURE EXTRACTION

Selected work

Ten activations from the last six years.
Hover a card to inspect it. Every visual is generated live — no images.

01 / PUBLISHED RESEARCHCNN · GAN · PyTorch

Automatic image colorization

A CNN that learns L→ab colour mappings, later extended with a GAN and served as an API. Published in the KIET Journal of Computing & Information Sciences (2021).

Deep learningGenerativePaper
02 / PRODUCTION PIPELINEFaceNet · ArcFace · OpenVINO

Face recognition pipeline

Detection, alignment, embedding and matching — FaceNet, ArcFace & CosFace inference plus RetinaFace on OpenVINO, built for Aletheia AI's product.

EmbeddingsEdge inference
03 / M.SC. THESIS · ONGOINGRPTU · PyTorch

3D human mesh recovery

Estimating 3D human pose and body shape from single images — adapting large-scale datasets to a new parametric body model and training multi-person recovery networks.

3D visionBody modelsDatasets
04 / GENERATIVE · APIGAN · TensorFlow · Keras

GAN image-colorization API

The published CNN, extended with a conditional GAN generator and served over an API. Separate generators trained for general scenes, coastlines and people, so grayscale input picks the domain expert.

GANsModel servingNotebooks
05 / RESEARCH

Egocentric grasp synthesis

Contextual reasoning for strain-aware hand–object grasps from a first-person view.

06 / MOBILE

Head-pose Android app

On-device head-pose estimation with threshold-triggered capture.

07 / EXPERIMENTS

OpenCV & image processing

Filters, Haar cascades, LBPH recognition and feature extraction — the fundamentals, by hand.

08 / CLASSIFICATION

CIFAR-10 classifier

A convolutional classifier across ten object classes — my first end-to-end training loop.

09 / SYSTEMSPython · ccxt · pandas

Momentum-breakout trading bot

Testnet-first crypto bot on 15-minute candles: EMA/RSI/volume entries, hard risk rails (2% risk cap, 20% notional cap, daily kill switch), exchange-side stops with software fallback, retry layer and fee-aware PnL journaling.

Risk engineLive systemsTestnet
10 / TOOLING · MINDGARAGEFlask · OpenCV · MMDetection

In-house Roboflow-style annotation app

A self-hosted dataset tool for the MindGarage lab: upload images or videos, extract frames, draw boxes in the browser, split train/test/validation and launch MMDetection training, all from one Flask app.

Data toolingDetectionWeb app
25+ repositories · Arctic Code Vault contributorOpen GitHub

LAYER 02 · EMBEDDING

Curious by default.
Rigorous by training.

I'm Saad — a computer-science M.Sc. student at RPTU Kaiserslautern and former research assistant at DFKI, Germany's national AI research centre.

My path runs from a bachelor's final-year project that became a published paper, through production face-recognition systems at Aletheia AI, to research on 3D human understanding today. What connects it all is a fascination with how machines perceive — and the discipline to make it work outside a notebook.

Off the keyboard I've captained a departmental football team, organised university events, and raised funds for children's education. Good work, I've learned, is rarely a solo effort.

Based in
Kaiserslautern, Germany
Focus
Computer vision · 3D humans · ML systems
Languages
English · Urdu · German (learning)
Open to
ML / CV engineering & research roles

drag to rotate · skills embedded in ℝ³

Toolkit

PythonPyTorchTensorFlowOpenCVNumPyPandasscikit-learnOpenVINOC / C++ / C#JavaScriptReactGitLinuxSQLAndroid

LAYER 03 · TRAINING LOG

Ten epochs of learning

Every role is an epoch. The curve is my learning rate.
Hover or tap a point to read the log entry.

loss (skills)val_loss (impact)
201620182020202220242026
epoch 10/10Oct 2022 — Sep 2023

Research Assistant — DFKI

Deutsches Forschungszentrum für Künstliche Intelligenz. Machine-learning pipelines, semantic datatype checking and dataset generation; built predictive models on large datasets to improve decision-making.

ML pipelinesLarge datasetsResearch
  1. OCT 2022 — SEP 2023

    DFKI · German Research Center for Artificial Intelligence

    Research Assistant

    Machine-learning pipelines, semantic datatype checking and dataset generation. Worked with large datasets to build predictive models that enhanced decision-making processes.

  2. NOV 2020 — MAR 2022

    Aletheia AI

    Machine Learning Engineer

    Joined as a system-deployment intern; promoted within months to a full-time role in product development. Core contributor to the facial-recognition pipeline.

  3. DEC 2020 — FEB 2021

    Amal Academy · Stanford-funded fellowship

    Career Prep Fellow

    Selected from 4,500+ applicants for a 150-hour programme in leadership, communication and problem-solving.

  4. 2016 — 2020

    Early internships

    AI · Web · IT

    Digital Landscape (AI intern — built a face-recognition system), Interns Pakistan (front-end), Pakistan Civil Aviation Authority (ASP.NET) and Pakistan Television (IT).

LAYER 04 · CHECKPOINT

The résumé, exported.

Three pages. Education, experience, projects, publications and certifications — the full checkpoint file.

Download PDF
Muhammad_Saad_Najib_CV.pdf · 194 KB · updated 2026
PDF · 3 pages

LAYER 05 · OUTPUT

Let's build something
that sees.

saad@portfolio ~ % contact --all

          

          
        

        Say hello
      

    

  


  

  
  
    
Muhammad_Saad_Najib_CV.pdf
Download

    
  

  
  
    
esc