Sneha Patel — Computer Science Student & AI/ML Developer Portfolio | Machine Learning, Computer Vision, Agentic AI & IoT
I am Sneha Patel, a Computer Science student at Navrachana University and an AI/ML developer. I build practical AI systems, agentic architectures, and edge solutions bridging computer vision, machine learning, and physical hardware.
I build practical AI systems at the intersection of machine learning, computer vision, agentic AI, and embedded hardware. From leading the AI architecture for Arcanum AI (6th Rank / 160 teams at TetraTHON 2026) to co-authoring edge smart waste segregation research for SustainX 2026, I focus on engineering systems that solve grounded, real-world problems.
expertise
meets
purpose
AI Systems & Applied Machine Learning
Designing, training, and deploying practical machine learning models tailored for real-world products and workflows.
Custom classification and regression architectures using PyTorch and Scikit-Learn. Transfer learning and fine-tuning convolutional networks and Vision Transformers. Rigorous model evaluation, validation metrics, and confusion matrix profiling. Decision intelligence ranking algorithms with explainable feature weights.
Agentic AI & LLM Integration
Developing intelligent conversational agents, knowledge-grounded RAG pipelines, and prompt orchestration engines.
Retrieval-Augmented Generation (RAG) referencing domain-specific documentation. Prompt engineering and contextual reasoning pipelines with structured outputs. Multi-agent task decomposition and asynchronous tool calling workflows. FastAPI microservice endpoints connecting LLM reasoning to frontend apps.
Computer Vision & Digital Forensics
Computer vision pipelines for object detection, produce sorting, explainable AI, and forensic image analysis.
Edge object detection and classification using YOLOv8 and ResNet architectures. Digital forensics feature extraction using Error Level Analysis (ELA) and DCT frequency analysis. Explainable AI visualization using Grad-CAM attention heatmaps. OpenCV image preprocessing, color segmentation, and background isolation.
IoT Prototyping & Edge Hardware
End-to-end hardware prototypes connecting microcontrollers, sensor telemetry, and on-device edge inference.
Raspberry Pi & Arduino Uno dual-architecture firmware and UART serial bridges. Sensor integration (ultrasonic, PIR presence, MQ gas, DHT environmental sensors). Actuator and servo motor calibration for automated physical mechanisms. Lightweight edge inference and MQTT cloud telemetry transmission.
Research & Publications
Investigating applied machine learning at the edge and longitudinal decision support architectures.
GreenSort: An Integrated IoT and Deep Learning System for Real-Time Municipal Solid Waste Segregation, Hazard Detection, and Route Optimization
Authors: Hetvi Sheth, Parisha Parmar, Sneha Patel
Venue: SustainX 2026 (Navrachana University)
Advisory: Prof. Gaurav Kumar Gaharwar, Assistant Professor, CSE, Navrachana University
Municipal solid waste (MSW) generation is projected to reach 3.4 billion tonnes annually by 2050. Existing smart bin deployments focus almost exclusively on fill-level monitoring, leaving waste classification, weight-based density anomaly detection, and hazardous gas monitoring largely unaddressed. GreenSort is a prototype Smart Waste Management System (SWMS) designed for municipal deployment. An Arduino Uno multi-sensor array feeds a Raspberry Pi running a custom-trained ResNet classifier fine-tuned via transfer learning, achieving 90 mAP at 1.7s edge inference latency on CPU. Telemetry and classification tags stream via MQTT to a Supabase cloud database, driving dynamic route optimization and citizen overflow reporting. Total prototype cost: approximately ₹5,000.
- Custom ResNet edge classifier achieving 90 mAP across 4 waste classes
- Edge inference latency of 1.7s on Raspberry Pi CPU hardware
- Integrated 6 sensors including ultrasonic (MAE 3.2%) and load cell (MAE 2.8%)
- Classification-aware dynamic route optimization for municipal trucks
- All 8 prototype test cases validated successfully
DermaJourney: Multi-Modal Longitudinal Decision Support Architecture for Chronic Dermatological Care
Authors: Sneha Patel
Venue: Undergraduate Academic Research Proposal
Investigates a multi-modal decision support framework for chronic dermatological conditions. Unlike conventional single-snapshot classifiers, DermaJourney explores sequential lesion feature delta tracking coupled with clinical NLP extraction from patient-reported symptom chronologies. Focuses on clinician-in-the-loop decision support with transparent uncertainty bounds, framed strictly as academic methodology research without unverified diagnostic claims.
- Longitudinal lesion change tracking framework across clinic visits
- Clinical NLP feature schema for patient symptom chronologies
- Clinician-in-the-loop decision support methodology
- Rigorous academic scope without unsubstantiated clinical deployment claims
“Her team delivered an outstanding context-aware developer copilot within 36 hours, distinguishing itself for technical depth, AST token parsing, and clean architecture across 160 teams.”
“The NeuraBin edge waste sorting prototype demonstrated impressive engineering practicality, achieving 90 mAP across 4 streams with a 1.7s cycle time within a ₹5,000 hardware budget.”
“Her graph database integration mapped sentiment patterns our standard NLP pipeline couldn't detect. The Neo4j context pathways added a new dimension to our risk assessment.”
“We were genuinely impressed by the team's presentation and analytical logic. The supplier ranking agent showcased impressive depth, earning our high commendation and appreciation.”
Technical Skills
Technologies I have actively built with across production prototypes, research, and hackathons.
Filter projects by skillProgramming Languages
Machine Learning & Deep Learning
Computer Vision
Agentic AI & LLM Applications
Data Science & Analytics
Backend & API Development
Databases
Cloud, Deployment & Hardware
Let's Build
Practical AI Systems
Whether you are looking for an AI/ML intern, a research collaborator, or an engineer to build production-grade agentic and vision systems — I bring research-grounded rigor and practical execution. Open to internships, research roles, and technical collaborations.
Connect & Collaborate
Have an internship opening, research opportunity, or technical challenge? I'm always glad to discuss AI architectures, model evaluations, and real-world system deployments.