› jenish_patel
Back home

All Projects

Full StackFeatured
2025

Auction Master

Full-stack auction platform with role-based access for admins, customer representatives, and buyers. Features live bidding with reserve-price enforcement, preference-based email alerts, image uploads, and an admin analytics dashboard.

3-role system · 20+ pages · preference-matched email notifications on new auctions

Node.jsExpressMySQLFirebase+4
AI / MLFeatured
2026

Edge AI LLM Benchmark

Browser-native LLM benchmarking tool that runs small open-source models entirely via WebGPU — zero server, zero API cost. Measures TTFT, tokens/sec, thermal throttling, and MMLU accuracy, then compares against cloud APIs (Gemini, Claude) on cost and quality.

Llama 3.2 1B: 73–172ms TTFT · 48–52 TPS · $0 vs. measurable cloud API cost per 20 queries

WebGPUWebLLMJavaScriptPython+3
AI / MLFeatured
2025

Industrial Fault Detection via SNNs

Brain-inspired computing research project comparing CNN and Spiking Neural Network architectures for anomaly detection on industrial bearing vibration data (CWRU) and cardiac ECG signals. Implements CNN→SNN weight transfer and surrogate gradient training across 3 architectures × 2 datasets.

6 notebooks: CNN→SNN weight transfer + surrogate gradient SNNs on CWRU bearing & MIT-BIH ECG

PyTorchTensorFlowSNNCNN+4
AI / ML
2025

Customer Churn Prediction

Explainable churn prediction pipeline trained on three real-world datasets — Telco (~7K rows), SaaS (~963 rows), and BankChurners (~10K rows). Compares Logistic Regression, Random Forest, and XGBoost with dataset-specific hyperparameter tuning, then explains every prediction via SHAP beeswarm and bar plots.

AUC ≥ 0.85 on Telco & BankChurners · SHAP beeswarm + bar plots per dataset per model

PythonXGBoostSHAPscikit-learn+2
AI / ML
2025

Belief-State Robot Localization

AI research project implementing robot self-localization under uncertainty across 4 phases: optimal BFS over belief-set state space → greedy strategy → CNN regressor trained to predict localization cost → 3-policy comparison. The agent navigates a maze without knowing its position, narrowing a belief set to a single cell.

CNN regressor (π₁) outperforms greedy baseline (π₀) · dual-policy π₂ evaluated across belief sizes 1–20

PythonPyTorchBFSCNN+3
Mobile
2023

Android Photo Album Manager

Native Android photo management app built in Java with 11 Activities covering the complete feature set: create/rename/delete albums, add and tag photos from device storage, search across all albums by Location and Person tags with autocomplete, slideshow view, and on-device serialized persistence.

11 Activities · AND/OR tag search with autocomplete · zero third-party dependencies

JavaAndroid SDKRecyclerViewMaterial Design+2
Systems
2023

Java Chess Engine

Fully playable two-player console chess engine in Java implementing the complete rule set: all 6 piece types with legal move validation, check and checkmate detection, en passant, castling with first-move tracking, pawn promotion to any piece, draw offer, and resign.

Full rule set: en passant · castling · checkmate detection · pawn promotion

JavaOOPGame LogicAlgorithms+1