all my
Projects
Sneha Patel - Projects
Dynamic Curated Case Studies Tailored for Placements
From deepfake detection to IoT waste sorting - each project represents a different problem I cared about solving.
Interactive Skills Mapping
Click any skill to instantly filter and showcase projects that use it.
Languages
Frameworks
Tools & Hardware
(01)
DeepShield AI Forensics
Computer Vision
A visual forensics pipeline I built to extract ELA and DCT compression artifacts, combined with a Vision Transformer to detect deepfakes. Includes an FGSM noise immunization mechanism.
Stack: PyTorch, FastAPI, Next.js, OpenCV, ViT, FGSM
Key Features: Combined custom digital forensics algorithms (ELA, DCT frequency) with a pre-trained Vision Transformer (ViT-Base-16) model · Built custom Grad-CAM visualization hooks to output pixel heatmaps of flagged anomaly areas · Designed a dual-pass verification system to verify fakes that easily bypass standard CNN model boundaries · Implemented a prototype FGSM noise-injector script to immunize original photos before web upload
Solution: I built a dual-pipeline forensic engine. It first extracts Error Level Analysis (ELA) and high-frequency Discrete Cosine Transform (DCT) artifacts, isolating compression discrepancies. It then feeds these artifacts into a Vision Transformer (ViT-Base-16) to classify the image. I also integrated an FGSM adversarial noise injection pipeline to pre-emptively immunize photos against facial cloning.
Impact: Achieved a 97.5% classification accuracy on unseen, highly compressed GAN holdout sets. Reduced false-positive rates on heavily artifacted social media images by 34% compared to standard spatial detectors.
(02)
Local Crop Classifier Study
Deep Learning
An agricultural computer vision study comparing six CNN architectures trained on locally captured vegetable datasets to optimize resource-constrained edge deployments.
Stack: PyTorch, ResNet18, Flask, OpenCV, Python
Key Features: Manually captured and preprocessed over 1,200 photos of regional vegetables (Karela, Parwal) in varying wholesale market lighting conditions · Evaluated accuracy and hardware latency trade-offs across 6 networks, including depthwise-separable designs · Wrote custom OpenCV filters to perform automatic background isolation and HSV color segmentation · Configured a Flask web interface to test real-time validation speeds and latency metrics
Solution: I manually captured and preprocessed 1,200+ photos of regional vegetables under varying wholesale market lighting conditions. I then compared six different convolutional network designs, evaluating lightweight depthwise-separable designs against heavier ResNet configurations. To ensure robustness, I built an OpenCV-based background segmenter and class-balancing filter to isolate organic structures in real-time.
Impact: Identified optimal network parameters that achieved 92% real-world accuracy while reducing inference latency by 45% compared to default ResNet configurations, making it viable for low-cost edge deployments.
(03)
Greensort: IoT Waste Sorter
IoT & Computer Vision
An edge-AI hardware prototype running a quantized YOLOv8 model on a Raspberry Pi 4, communicating sorting signals to an Arduino-driven mechanical flap.
Stack: YOLO, OpenCV, Python, Raspberry Pi, Arduino
Key Features: Programmed serial communication (Serial over USB) to link the Raspberry Pi 4 vision unit with the Arduino Uno motor controller · Quantized YOLOv8 weights to INT8 to run at 12fps inference speed directly on local edge hardware cores · Designed and calibrated a mechanical sorting gate powered by a high-torque MG996R servo motor · Integrated HC-SR04 ultrasonic distance sensors to monitor bin capacities and send telemetry alerts
Solution: I engineered a self-coordinating hardware and AI prototype designed to classify and sort garbage at the point of disposal completely offline. A Raspberry Pi runs a highly quantized YOLOv8 model, processing frames in <80ms. It communicates sorting decisions to an Arduino Uno over a USB Serial link, driving physical servo flaps to segregate recyclables, organic matter, and hazardous items.
Impact: Successfully sorted local plastic and organic waste samples in real-time with a total loop latency (camera capture to physical flap movement) of under 180ms.
(04)
Logistics Supplier Ranker
Data Analytics
A multi-variable regression model built during a hackathon to score and rank cargo providers based on defect rates, costs, and late deliveries.
Stack: Python, Scikit-Learn, Regression Models, Pandas
Key Features: Fitted a custom multi-variable regression algorithm weighting late-deliveries, defect logs, and cost metrics · Designed a data validation and cleansing pipeline to ingest sparse logistical records using pandas · Built an analytical prototype dashboard in Streamlit to visualize ratings and flag outlier suppliers · Earned direct commendation from the CEO of Mesh Works during the hackathon presentation for ranking accuracy
Solution: I constructed a decision intelligence agent that automates risk profiling. By ingesting sparse logistical records using Pandas, the system fits a customized multi-variable regression model weighting late-deliveries, defect logs, and cost metrics to score each provider. I visualized the anomalies in a Streamlit prototype.
Impact: The regression coefficients successfully highlighted anomalous suppliers, earning direct commendation from the CEO of Mesh Works during the hackathon presentation for its immediate practical applicability.
(05)
NLP Crisis Context Mapper
Machine Learning & NLP
An NLP crisis evaluation tool built during my CSRBOX internship that maps behavioral text statements and triggers into a Neo4j graph database.
Stack: Python, Scikit-Learn, Neo4j, NLP, Pandas
Key Features: Wrote text preprocessing scripts utilizing TF-IDF, tokenizers, and sentiment analyzers in Python · Designed a Neo4j graph schema to link nodes representing patient IDs, sentiment scores, and distress tokens · Trained a Scikit-Learn ensemble model to categorize statements into clinical risk brackets · Wrote Cypher queries to help evaluators trace chronological context pathways instead of isolated keywords
Solution: I developed an NLP engine that predicts crisis risk levels from user statements. Instead of just using TF-IDF vectors, I utilized a Neo4j Graph Database to map connections between emotional states, anxiety triggers, and temporal behavioral changes. A Scikit-Learn ensemble model categorizes the statements into clinical risk brackets.
Impact: Enabled clinical evaluators to trace chronological context pathways, significantly enhancing pattern detection and reducing false-positive risk alerts by providing graph-backed contextual evidence for every prediction.