You will get Classify images with deep learning models and computer vision
Project details
Most ML freelancers hand you a model that works in a notebook and falls apart in production. I build lean, efficient models — grounded in Green AI research — that deliver strong accuracy without bloated compute costs, and come with the documentation your team actually needs to use them. Backed by published research in computer vision and IoT systems, plus a decade of translating complex technical work into clear, actionable results.
Machine Learning Tools
ChatGPT, Keras, NumPy, OpenCV, pandas, Python Scikit-Learn, PyTorch, scikit-learn, TensorFlow, XGBoostWhat's included
| Service Tiers |
Starter
$75
|
Standard
$200
|
Advanced
$450
|
|---|---|---|---|
| Delivery Time | 3 days | 7 days | 14 days |
Number of Revisions | 1 | 2 | 4 |
Number of Model Variations | 1 | 2 | 4 |
Number of Scenarios | 1 | 3 | 5 |
Number of Graphs/Charts | 2 | 5 | 10 |
Model Validation/Testing | - | ||
Model Documentation | - | ||
Data Source Connectivity | - | - | |
Source Code |
Optional add-ons
You can add these on the next page.
Additional Revision
+$20
Additional Model Variation
(+ 1 Day)
+$20
Additional Graph/Chart
(+ 2 Days)
+$30
Model Validation/Testing
(+ 2 Days)
+$20
Model Documentation
(+ 1 Day)
+$20About Kishwar
Computer Vision & Edge AI Researcher | YOLO, MobileViT, CNN
Dera Ghazi Khan, Pakistan - 9:17 pm local time
I design lightweight computer vision and IoT models that run on constrained hardware —
not just in a notebook. My background is applied research in "Green AI": building models
that are accurate enough to matter and small enough to actually deploy, without the
compute bill of a full-scale architecture.
WHAT I BRING
I'm currently completing an MSIT (Deep Learning & Vision Models specialization) at Ghazi
University, where my thesis work centers on lightweight Vision Transformer architecture
for image classification on resource-constrained systems. I've also published peer-reviewed
research on IoT sensor systems and cloud-integrated safety monitoring. Alongside this,
I've spent 10+ years as a technical instructor — which means I don't just build models,
I can explain exactly what they're doing and why, to technical and non-technical
stakeholders alike.
CORE CAPABILITIES
▸ Computer Vision — image classification & object detection (MobileViT, MobileNetV2/V3,
Swin Transformer, ResNet50/101, EfficientNet, YOLO, SSD)
▸ Edge AI / TinyML — model design and evaluation for constrained hardware (ESP32-class
devices), transfer learning, feature extraction
▸ IoT & Cloud Data Pipelines — sensor-to-cloud architectures using MQTT, real-time
streaming data, automated threshold-based alerting
▸ Classical ML & Statistical Modeling — regression (Linear, Polynomial, Random Forest,
SVM, Ridge/Lasso), classification (XGBoost, KNN, Logistic Regression, Naive Bayes,
Decision Trees), ensemble methods (Bagging, Boosting, AdaBoost)
▸ Data Engineering — dataset collection, cleaning, augmentation, SMOTE oversampling,
cross-validation, ROC-AUC / confusion matrix / error-profile evaluation
▸ Technical & Research Writing — peer-reviewed publication experience, structured
documentation, clear reporting for non-technical audiences
VERIFIED WORK
▸ Built and curated a 4,900-image dataset across 10 flower categories for a lightweight
MobileViT classification model, combining original data collection with the public
Oxford 102 dataset to improve coverage — MSIT thesis research (2024–2026, in progress)
▸ Co-developed and published an ESP32-based dual-sensor gas safety system achieving a
verified 95%+ detection rate at hazard thresholds, using MQTT for low-latency alerting
— published in the International Journal of Innovations in Science & Technology (2025)
▸ Published a comparative methodology study on sentiment-based mental health monitoring
models — Spectrum of Engineering Sciences (2025)
▸ Recognized for scientific peer-review quality — Certificate of Excellence in Scientific
Reviewing, Asian Journal of Geographical Research (2025–2026)
▸ 10+ years as a technical instructor with a maintained 100% regional board exam pass
rate — direct evidence of the communication and reliability I bring to client work
HOW I WORK
I'm early on Upwork but not early in the field — my research and technical background
translate directly into practical delivery. I'm upfront about scope: for production
deployment work, I'll tell you clearly if something is within my proven experience or if
it's a capability I'm extending into, so you're never surprised mid-project.
Message me or send an invite with a bit about your use case — I'll respond with a
straight assessment of fit before you spend a connect.
Steps for completing your project
After purchasing the project, send requirements so Kishwar can start the project.
Delivery time starts when Kishwar receives requirements from you.
Kishwar works on your project following the steps below.
Revisions may occur after the delivery date.
Data Preparation
I clean, preprocess, and validate your data (handling missing values, imbalance, augmentation) to ensure the model is trained on solid foundations.