Freelancers using PyTorchFreelancers using PyTorch
Senior Software & ML Engineer | Zero to One Product Builder
    $•••••
    Earned
    6x
    Hired
    5.0
    Rating
    72
    Followers
Senior Software & ML Engineer | Zero to One Product Builder
AI Agents | LLMs, Computer Vision & Full-Stack Dev
    92
    Followers
AI Agents | LLMs, Computer Vision & Full-Stack Dev
Cover image for 🌳 Tree Crown Detection, Done Right We built an AI system th...
🌳 Tree Crown Detection, Done Right We built an AI system that detects and segments individual tree crowns in aerial imagery, producing clean, non-overlapping boundaries. Using RetinaNet + Mask R-CNN, our model outperforms Detectree2 with fewer false positives, no over-segmentation, and GIS-ready outputs, even in dense forests. #AI #ComputerVision #RemoteSensing
19
497
Enterprise Cloud · DevOps · AI/ML · Email · Marketing Expert
    $10k+
    Earned
    13x
    Hired
    4.6
    Rating
    46
    Followers
Enterprise Cloud · DevOps · AI/ML · Email · Marketing Expert
Cover image for I’m Usama Idrees, IT &
I’m Usama Idrees, IT & AI Specialist based in Islamabad, Pakistan. Thrilled to be part of this amazing community and to connect with talented Professionals from around the world! 🚀 A little about me: What I do: I lead tech and product strategy, focusing on building innovative solutions that empower businesses and individuals. Passion: Creativity meets technology — I love turning ideas into impactful products. Looking for: Collaborations with designers, developers, and creative minds who want to build something extraordinary. Excited to learn, share, and grow together here. If you’re working on something cool or looking for a tech partner, let’s connect! 💡 What’s your current creative project? Drop it below — I’d love to check it out. 👇
2
30
949
AI/ML Engineer crafting intelligent systems & AI solutions.
    15
    Followers
AI/ML Engineer crafting intelligent systems & AI solutions.
Cover image for 1. What it is: An
1. What it is: An advanced e-commerce web app that combines Graph Neural Networks (GNNs) and Generative AI to identify "toxic inventory"—products, sizes, or suppliers driving high customer returns. 2. GNN Predictive Modeling: Built a heterogeneous GNN using PyTorch Geometric to model relationships between customer demographics and product features to predict future return risks. 3. AI Attribute Enrichment: Automated a metadata extraction pipeline using the Gemini API with local caching to parse raw HTML product descriptions into structured product features (fabric, fit, pattern). 4. Interactive Dashboard & Reporting: Developed a Streamlit dashboard with Supabase Auth and integrated ReportLab to generate boardroom-ready PDF return audit reports. 5. Tech Stack: Python, PyTorch Geometric, Streamlit, Gemini API, Supabase, ReportLab.
0
88
Cover image for Scientific Image Forgery Detection — Kaggle Competition Part...
Scientific Image Forgery Detection — Kaggle Competition Participated in the ongoing Kaggle competition on Copy-Move Forgery Detection in Scientific Images, aimed at identifying manipulated biomedical figures that can compromise research integrity. For this challenge, I developed a ResNet50 + U-Net hybrid segmentation model using PyTorch, designed to detect and segment forged regions at the pixel level. My approach combines Dice and Focal losses for balanced training, WeightedRandomSampling to oversample forged images, and Test-Time Augmentation (TTA) to improve prediction robustness. Achieved an initial score of 0.303 on the public leaderboard. I’m continuing to experiment with architecture tuning, learning rate schedules, and other loss functions to further enhance performance and generalization.
0
146
Cover image for I recently fine-tuned the Mistral 7B Instruct model on a dat...
I recently fine-tuned the Mistral 7B Instruct model on a dataset of NDA (Non-Disclosure Agreement) documents — building an AI reviewer capable of identifying compliance issues and clause inconsistencies. To make the model more accessible, I converted the trained weights to CPU-compatible files, allowing efficient inference without GPU requirements. Model: Mistral 7B Instruct v0.1 Focus: Legal text review & semantic understanding Tech: PyTorch, Transformers, Kaggle Check out the full notebook here → [Kaggle Project Link (https://www.kaggle.com/code/karansingh123456/nda-reviewer-model-training)] My Kaggle account here → Profile (https://www.kaggle.com/curiouscyborgs) #AI #NLP #SentimentAnalysis #DeepLearning #CNN #LSTM #DataScience #GitHub
0
82
Cover image for In this project, I developed a Sentiment Analysis Web App us...
In this project, I developed a Sentiment Analysis Web App using deep learning (CNN) and traditional models to classify text sentiment with high accuracy. The system includes a complete evaluation pipeline comparing CNN, LSTM, Logistic Regression, Random Forest, and Naive Bayes — analyzing performance across multiple iterations and datasets. Key Highlights: Built a Streamlit-based web app for real-time sentiment classification Developed and evaluated multiple models for accuracy and F1-score Created detailed analysis reports and prototype schematics Project here → GitHub Repository (https://github.com/Imkaran04/Sentiment_Analysis_Web_App/tree/main) Reports: Sentiment Analysis Report (PDF), Product Prototype Diagram Tech Stack: Python, Streamlit, TensorFlow/Keras, Scikit-learn, Matplotlib
0
118
AI & ML Engineer|Real-Time Computer Vision & Edge AI Expert
AI & ML Engineer|Real-Time Computer Vision & Edge AI Expert
Cover image for AI Shoplifting Detection System: Intelligent
AI Shoplifting Detection System: Intelligent Video Analytics for Retail Loss Prevention Protect your storefront with an automated AI security solution that never sleeps. This project implements a full-stack Computer Vision pipeline capable of monitoring 16+ simultaneous RTSP streams. By utilizing ByteTrack for stable person re-identification and Deep Learning action classifiers, the system detects unauthorized entry into staff zones and alerts management to shoplifting incidents as they happen. A robust, scalable solution for grocery stores and retail outlets looking to modernize their security infrastructure.
0
134
Cover image for AI-Powered PPE Compliance & Safety
AI-Powered PPE Compliance & Safety Monitoring
0
128
Cover image for multi camera ai tracking for
multi camera ai tracking for mask detection and motion analytics Stop monitoring manually. Start automating safety. In high-stakes environmentshospitals, construction sites, and manufacturing plantscompliance isn't optional. I build Industrial-Grade AI Video Analytics that combine real-time Face Mask Detection with advanced Person Movement Tracking to ensure 24/7 safety oversight without human error. The Synergy: Why Both Matter Most developers offer one or the other. I integrate them into a single, high-performance pipeline: Compliance Monitoring: Instant detection of PPE/Face Mask violations with timestamped logging. Behavioral Tracking: Beyond simple detection, I track individual movement paths to identify "High-Risk" behaviors or unauthorized entry into restricted zones. RTSP Scalability: My systems don't just work on one webcam; they are optimized to handle multi-camera RTSP feeds (16+) with zero lag. Key Features of the System: Dual-Stream Intelligence: Real-time Mask/No-Mask classification paired with unique Person IDs (Re-ID). Zone-Aware Analytics: Define specific "Mask-Mandatory Zones" vs. "Common Areas" to reduce false alerts. Motion & Velocity Insights: Track if a person is running, loitering, or entering a hazardous area without prop
0
140
Cover image for yolov11 alpr anpr system with
yolov11 alpr anpr system with rtsp stream and sql database You will get a production-ready AI-powered vehicle monitoring and license plate recognition system designed for real-world environments such as apartments, parking garages, and secure facilities. This solution combines advanced computer vision (YOLO + OCR + tracking) with a modern React dashboard to deliver accurate, real-time vehicle identification and tracking. What sets my work apart is the focus on reliability, privacy, and scalability. The system is optimized to reduce OCR errors using intelligent plate stabilization, ensuring consistent and accurate results even in challenging conditions. All data is processed and stored locally, making it ideal for privacy-sensitive deployments. The dashboard provides live vehicle cards with owner details, parking status, and access control (authorized, visitor, blocked), giving you full visibility and control. The system is modular and can be extended with features like automated gate control, alerts, analytics, and multi-location support. With a strong background in AI and computer vision systems, I build solutions that are not just demos—but ready for real-world production use.
0
128
I’m an AI & Machine Learning engineer with expertise in deve
    6
    Followers
I’m an AI & Machine Learning engineer with expertise in deve
Cover image for What your attention heatmap isn't
What your attention heatmap isn't telling you Everyone's staring at attention heatmaps and calling it "interpretability." Almost nobody's asking whether a single attention map actually tells you what the model used to make its decision. It doesn't. Not on its own. A raw attention map from layer 8 shows you what layer 8 attended to. It says nothing about how that signal got mixed, diluted, or overwritten by every layer before and after it. Attention rollout fixes this — and I built a walkthrough to show why it matters. Here's what makes it more than a "pretty heatmap" demo: Instead of visualizing one layer's attention, I traced how information actually flows through the full transformer stack. → Every layer's attention matrix is extracted, per head, per token → Multi-head attention is averaged, then combined with the residual connection (identity + attention) — this is the step most tutorials skip, and it's the one that actually matters → The combined matrices are matrix-multiplied layer by layer, rolling attention forward from input to output → The result: a single map showing genuine token-to-token influence across the entire network, not just one layer's snapshot The overlay shows you everything: → Per-layer attention vs. rolled-out attention, side by side → Token importance scores overlaid directly on the input text → A comparison view: which tokens raw attention says "matter" vs. which ones rollout says actually matter → Head-level breakdown so you can see which heads specialize vs. which are noise No black box. No "trust me, the model looked here." Just linear algebra, applied honestly across every layer instead of cherry-picking one. Built with PyTorch + HuggingFace Transformers + Matplotlib. Runs on any pretrained transformer, fully offline. ⚠️ Important: attention rollout is an approximation, not ground truth. It assumes attention is the primary information pathway, which ignores MLP layers and can still mislead for very deep models. Treat it as a debugging lens, not proof of causality.
0
205
Cover image for Everyone's racing to add biometrics
Everyone's racing to add biometrics to logins. Almost nobody's asking what happens when you can't — or shouldn't — touch the sensor. Shared kiosks, clinical settings, accessibility needs, hygiene-sensitive environments. Fingerprint readers and face unlock assume contact or a stored faceprint. Sometimes you want authentication that touches nothing and stores no biometric image of you at all. So I built GestureAuth — a contactless authentication system where your "password" is a sequence of hand gestures performed in front of a standard webcam.
1
186
Cover image for Everyone's talking about quantum computing.
Everyone's talking about quantum computing. Nobody's using it to feed farmers. India loses 20–30% of its crop yield every year to diseases and pests. Not because farmers don't care — but because early detection is hard, expensive, and inaccessible to the people who need it most. The existing solutions? Either a basic image classifier trained on lab-perfect photos that fail in real field conditions, or an agronomist visit that costs time and money most small farmers don't have. So I built QuantumEdge AgriGuard — a hybrid Quantum Neural Network app where a farmer can photograph a diseased leaf on their phone and get an instant diagnosis in under 5 seconds. Here's what makes it different from just another plant disease detector: Instead of a pure classical CNN, I built a hybrid architecture — a ResNet/EfficientNet backbone extracts visual features, then passes them into a Variational Quantum Circuit (VQC) for the final classification. The quantum layer uses angle embedding + StronglyEntanglingLayers, which gives it a measurable edge on small, noisy datasets — exactly the kind of data you get from Indian field conditions. The app doesn't just tell you what disease it is. It gives you: → Confidence score → Organic + chemical remedies (India-specific) → Yield impact estimate → A live classical vs quantum accuracy comparison so you can see the difference yourself I tested the quantum advantage claim honestly — ran both models on the same downsampled PlantVillage dataset and tracked accuracy, F1-score, and inference time side by side. The results are on the dashboard. No hand-waving. Built with PennyLane + PyTorch + Plotly Dash. Designed to run on simulators today and on QpiAI-Indus 25-qubit hardware tomorrow.
1
0
216
Cover image for  A Neural Network Visualization
A Neural Network Visualization Tool that demystifies AI! I built an interactive web application using Flask and PyTorch that doesn't just recognize handwritten digits it shows you exactly how the AI "thinks." When you draw a digit (0-9) on the canvas, the app processes it through a Convolutional Neural Network and generates a real-time visualization of every layer: from edge detection in the first convolutional layer, through pattern recognition, pooling, and feature extraction, all the way to the final classification. The tech stack includes Python, Flask, PyTorch, and vanilla JavaScript for the frontend. What makes this unique is the educational aspect each layer's activations are visualized using matplotlib, showing the 32 filters in Conv Layer 1, the 64 filters in Conv Layer 2, and the 128-neuron fully connected layer.
1
106
Senior AI/ML Engineer & Academic Mentor | Ph.D. (USTC)
    $1k+
    Earned
    1x
    Hired
    4
    Followers
Senior AI/ML Engineer & Academic Mentor | Ph.D. (USTC)
Founder & AI Systems Architect
New to Contra
Founder & AI Systems Architect
Cover image for Enterprise Multimodal AI Platforms (NDA)
Designed
Enterprise Multimodal AI Platforms (NDA) Designed and developed two production-grade enterprise AI platforms under NDA, taking ownership of the complete AI lifecycle from research and system architecture to production deployment and long-term platform support. The platforms integrated LLMs, RAG, Computer Vision, OCR, intelligent document processing, AI agents, custom neural networks, benchmarking, model evaluation, dataset engineering, synthetic data generation, model training, fine-tuning, optimization, and scalable production inference pipelines.
0
47
Cover image for Horse Health AI is a
Horse Health AI is a Golden Dragon AI DeepTech project for portable, non-invasive, multimodal animal health screening and longitudinal health intelligence. I designed the system around a compact multimodal sensor architecture combining RGB Computer Vision, multi-wavelength NIR, radiometric thermal imaging, structured/polarized optical sensing, and motion and geometry data. Rather than analyzing each signal independently, the AI architecture is designed to fuse complementary modalities and evaluate visible anatomy, gait and behavior, thermal patterns, vascular and perfusion-related features, tissue-related optical signals, movement, and individual history. A key part of the architecture is the Individual Digital Baseline. Each animal can become its own longitudinal reference, allowing new scans to be compared not only with validated population-level knowledge but also with that animal's historical normal patterns. The software architecture includes synchronized acquisition, calibration, multimodal registration, per-modality AI models, longitudinal baseline modeling, multimodal embeddings and fusion, veterinary knowledge/RAG, explainable anomaly localization, confidence and uncertainty assessment, and mobile and research interfaces. The project is designed as a screening and decision-support system, not as a replacement for veterinary diagnosis. Veterinary expertise, clinically confirmed data and validation are explicit parts of the development roadmap. The initial platform is focused on horses, with potential expansion to other animal species following species-specific validation.
1
108
Cover image for Golden Dragon Quantum Trading
AI-Native Multi-Agent
Golden Dragon Quantum Trading AI-Native Multi-Agent Intelligence for Active Crypto Trading Golden Dragon Quantum Trading is a highly customizable AI-native crypto trading intelligence and controlled execution platform. I designed the platform around a distributed multi-agent architecture: the trader communicates with one Main AI, while a network of specialized intelligent AI agents analyzes different market domains in parallel, including market structure, order books and liquidity, derivatives, volume and flow, whale activity, cross-exchange conditions, news, risk, and trading strategies. The architecture combines personalized trading strategies, manipulation and anomaly detection, whale and hidden-liquidity intelligence, multi-agent confirmation, hybrid quantum-classical computation, and a proprietary orchestration and real-time communication architecture. A separate System Control Core supervises system state and permissions, while Protective Gates validate risk, strategy, market, exchange, and execution conditions before authorized actions reach the Execution Robot. The platform is designed for multi-exchange operation, including Binance, Bybit, Bitget, OKX and Kraken. User funds remain in the trader’s own exchange accounts. I also built the customer-facing product environment, including registration, authentication, subscriptions, payments, personal access tokens, exchange connection, multi-exchange selection, TradingView integration, multilingual functionality, and the Main AI trading workspace. My role: Founder, AI Systems Architect & Developer Project: Architecture, AI/ML, Multi-Agent Systems, Trading Intelligence, Backend, Cloud, Product Development
0
51
Cover image for Project: Dragon Vision — Agent-Built
Project: Dragon Vision — Agent-Built AR Ingredient Scanning Effect Instagram Reel: https://www.instagram.com/reel/DeYMiNXsRBz/ invideo Project Link: https://invideo.io/share/Export-10-Oct-25thLT4QVMM Primary Effect: Custom AR-style scanning effect using AI-generated visual layers, scanning highlights, and animated information overlays. Process Note: I created Dragon Vision using invideo Editor's AI editing agents. Starting with AI-generated source footage, I directed the agent to build an AR-style ingredient scanning effect with animated overlays, scanning lines, and information panels. I refined the timing, opacity, and animation of the layers through the editing workflow. My screen recording demonstrates the agent workflow, editable layers, refinements, and my voice commentary.
0
81