{Erikas Kadiša}

Feb 2026 – May 2026 · 4 months

A TikTok-style mobile platform connecting startup founders with investors through short-form pitch videos. Designed the full system architecture in Figma and implemented core product flows: email/password and Google OAuth2 authentication with PKCE, vertical video feed with controller windowing and preloading, server-side search across startup titles and founder names, and a founder verification wizard with document upload.

FLFlutterDartSupabaseCloudflare R2+2

Feb 2026 – May 2026 · 4 months

A self-contained multi-hop LoRa mesh network for wireless command and telemetry of a physical rover. Built on XIAO microcontrollers running CircuitPython with a custom SX1262 driver, Bellman-Ford distance-vector routing, adaptive beaconing, and a store-and-forward DTN queue. Controlled entirely from a browser dashboard that connects over BLE — no servers, cloud, or external dependencies required.

PythonESP32LoRaWANBellman-Ford Routing+3

Sep 2025 – Jan 2026 · 4 months

A BiLSTM-CNN forecasting model for next-day closing price prediction on the GLD ETF. Aggregated 10 years of ETF trading data, engineered financial features including rolling volatility, moving average crossovers, and volume-weighted price indicators. Validated using expanding-window cross-validation, achieving +19.9% annualised simulated returns against a buy-and-hold baseline.

PythonJavaPyTorchNumPy+1

Mar 2025 – Dec 2025 · 10 months

Research assistant at the University of Luxembourg developing CNN-based classifiers for AI-generated code detection. Built a system that maps Java source code to 224×224 RGB images via binary-to-pixel encoding and distinguishes human-written code from outputs of five sources — ChatGPT (GPT-4), Claude Sonnet 4, Gemini 2.5 Flash, DeepSeek V3, and human — achieving 97.7% accuracy on a 67,760-sample dataset using DenseNet121, ResNet50, and VGG16 architectures.

PythonTensorFlowPyTorchCNNs