High-performance network tracking utility designed to monitor and display active connection telemetry inside the system notification register.
- Renders active download/upload telemetry directly inside the device status bar.
- Tracks Wifi vs Mobile connection data history dynamically, maintaining local records.
- Implements optimized battery-friendly background task polling to prevent OS throttling.
Tech Stack:
Java
Android SDK
SQLite
Broadcast Receivers
Techniques:
Background Service Polling
System Overlay Notifications
Local Database Logging
Security-focused storage vault masquerading behind a fully functional calculator interface.
- Encrypts local media storage layers dynamically, bypassing default device gallery indexes.
- Triggers intrusion monitoring logs (capturing device snapshots on invalid entry submissions).
- Applies device-level masking, hiding launcher components from active application lists.
Tech Stack:
Kotlin
Android Jetpack
Room Database
Crypto APIs
Techniques:
AES-256 Media Encryption
Biometric Access APIs
App Component Masking
Natural Language Processing binary classifier engineered to recognize duplicate questions to prevent semantic text redundancy.
- Extracted text similarity attributes including TF-IDF metrics, fuzzy ratios, and overlap statistics.
- Calculated cosine distance parameters across question pairs to identify sentence equivalence.
- Optimized binary XGBoost classifiers to classify matching pairs, lowering search redundancy.
Tech Stack:
Python
NLTK
Scikit-learn
XGBoost
Techniques:
Fuzzy String Matching
TF-IDF Vectorization
Cosine Similarity Distance
Classification model designed to categorize genetic mutation patterns based on scientific literature text summaries.
- Processed dense clinical paper texts, linking genetic patterns with target medical classes.
- Designed custom text sanitizers and high-dimensional TF-IDF pipelines to extract text features.
- Optimized multi-class Logistic Regression algorithms using log-loss evaluation statistics.
Tech Stack:
Python
Scikit-learn
NLTK
Pandas
Techniques:
Clinical Text Tokenization
Log-Loss Tuning
TF-IDF Feature Mapping
Graph link prediction system identifying missing edges on large-scale social network maps.
- Engineered graph metrics including PageRank, Adamic-Adar index, Jaccard similarity, and HITS scores.
- Decomposed link structures using SVD matrix operations to represent low-dimensional graph nodes.
- Trained classifier configurations using XGBoost to predict probable future connection recommendation lists.
Tech Stack:
Python
NetworkX
Scikit-learn
XGBoost
Techniques:
Graph Node Link Prediction
SVD Matrix Decomposition
Adamic-Adar / PageRank
Spatio-temporal regression model forecasting local passenger pick-up demand across NYC coordinate clusters.
- Clustered geographical sectors dynamically using K-Means spatial coordinate clustering.
- Formed rolling lag variables to capture time-series and seasonal demand frequency.
- Trained regression networks using Random Forests and XGBoost models to predict pickup densities.
Tech Stack:
Python
Scikit-learn
XGBoost
K-Means
Techniques:
Spatial Cluster Segmentation
Temporal Demand Modeling
Time-Series Regression
Multi-label text classification classifier designed to auto-assign tech tags based on questions textual content.
- Designed tag mapping pipelines based on Binary Relevance and SGDClassifier methods.
- Extracted structural features from combined conversational text and markdown code blocks.
- Configured training models using TF-IDF n-grams to label multi-target profiles.
Tech Stack:
Python
Scikit-learn
TF-IDF
SGDClassifier
Techniques:
Multi-Label Classification
Binary Relevance Method
TF-IDF Text Parsing
Classification system evaluating node telemetry states to predict probability of machine malware infections.
- Managed scaling operations on high-dimensional device logs and hardware parameter registers.
- Structured model parameters to handle target class imbalance distributions.
- Implemented decision architectures utilizing LightGBM and XGBoost, sorting node correlations.
Tech Stack:
Python
LightGBM
XGBoost
Random Forest
Techniques:
High-Dimensional Feature Map
Imbalanced Class Training
Gradient-Boosted Decision Trees
Autonomous steering engine leveraging behavioral cloning strategies to direct vehicles around course maps.
- Cloned NVIDIA autonomous models using an end-to-end Convolutional Neural Network (CNN).
- Designed image augmentation sets including Hue modifications, shifts, and horizontal flips.
- Integrated model parameters to issue steering predictions dynamically in simulator environments.
Tech Stack:
Python
TensorFlow
Keras
OpenCV
Techniques:
NVIDIA CNN Behavioral Cloning
Real-Time Image Augmentation
Computer Vision Preprocessing
Classification system processing temporal smartphone coordinates to classify client activity configurations.
- Modeled time-series coordinate telemetry using Long Short-Term Memory (LSTM) models.
- Structured filters to clean accelerometer/gyroscope signals and run Fourier transformations.
- Processed coordinate flows into temporal slices to train deep multi-class classification networks.
Tech Stack:
Python
TensorFlow
Keras
NumPy / Pandas
Techniques:
Recurrent LSTM Networks
Signal Preprocessing Filters
Fourier Transform Spectral Map
Generative recurrent network modeling piano themes to compose original classical music MIDI sequences.
- Transcoded musical notes and chord changes into dynamic categorical indexes using Music21.
- Engineered sequence prediction systems utilizing stacked Recurrent LSTM layers.
- Configured generation parameters using temperature sampling methods to regulate output creativity.
Tech Stack:
Python
TensorFlow
Keras
Music21
Techniques:
Sequential Recurrent Models
MIDI File Tokenization
Temperature-guided Composition
Lightweight card game published on Facebook Instant Games, optimized for instant load times and fast canvas redraw cycles.
Tech Stack:
HTML5
Javascript (ES6)
CSS3 Canvas
FB Instant SDK
Techniques:
UI Canvas Redraw Loop
Instant Preload Cache
State Management Routing
Localized Wireless Internet Service Provider (WISP) distribution network serving 140+ active concurrent subscribers.
- Deployed point-to-multipoint (PtMP) RF propagation links utilizing high-gain TP-Link Pharos CPE sector arrays, routing paths across high-loss line-of-sight zones.
- Configured dynamic Layer 3 routing protocols (OSPF multi-area) to guarantee path redundancy and sub-millisecond network convergence.
- Architected dynamic bandwidth allocation configurations using Hierarchical Token Bucket (HTB) Queuing and QoS profiles inside MikroTik RouterOS.
- Hardened access networks by deploying MAC bridge filters, static ARP table bindings, PPPoE tunnels, and strict client isolation policies.
Tech Stack:
RouterOS (MikroTik)
Pharos CPE (TP-Link)
Python API
OSPF
Techniques:
Hierarchical Token Bucket (HTB)
RF Path Propagation Analysis
MAC Bridge Security Hardening
Docker-orchestrated local server setup containerizing applications, reverse-proxy structures, and system configurations.
- Containerized local utility tools, databases, and microservices using Docker and Docker Compose.
- Configured Nginx Reverse Proxy routing layers to secure local ports and configure routing rules.
- Orchestrated telemetry monitors to inspect container logs, health status, and CPU thresholds.
Tech Stack:
Docker
Docker Compose
Linux (Ubuntu)
Nginx Proxy
Techniques:
App Container Isolation
Reverse Proxy Routing
Docker Bridge Networking
