Jonathan

Jesni

Modified Double U-Net

Dual-stacked U-Net for 3-class medical image segmentation (Background, Benign, Malignant). Uses an Ensemble Encoder fusing VGG-19, DenseNet-121, and Xception backbones with Softmax-based Attention Gates to route spatial cues. Optimized via AMP and a combined Cross-Entropy & Dice loss to handle class imbalance, achieving a ~0.85 Validation F1-Score and accelerating training by ~25%.

• Deep Learning
• Medical Imaging
• Computer Vision

BandWidth

Orchestrated a 5-agent cross-model pipeline (GPT-4o + DeepSeek-V4-Pro) across 6 containerized microservices to autonomously review, test, fix, and document GitHub PRs via the Band collaboration platform. Built a Flask webhook engine intercepting PR events, executing sandboxed pytest validations, and pushing autonomous fix commits via the GitHub REST API. Deployed on Google Cloud with 100% reliable agent handoffs and deterministic state-routing.

• Agentic AI
• Multi-Agent
• CI/CD

SynthRescue

Autonomous AI triage and synthetic data engine. Engineered a procedural 3D Blender pipeline to automate bounding-box annotations across occluded disaster scenes. Trained a custom YOLOv8 model on ~6,115 images to achieve 96.7% precision, and deployed a low-latency model inference endpoint (FastAPI REST, Docker, GCP) integrating Gemini AI to translate live drone telemetry into actionable triage reports within ~4.5 seconds.

• Computer Vision
• Synthetic Data
• Deep Learning

Ludex

A hybrid recommendation engine combining content-based filtering and collaborative filtering to improve relevance by ~12-18% over standalone baselines. Integrates diversity-aware re-ranking to increase catalog coverage by ~20% (reducing popularity bias) and handles cold-start user scenarios using metadata-driven fallback logic, evaluated on large-scale Steam interaction data.

• Machine Learning
• Recommendation Systems
• Python

File Converter

A deterministic two-pass document conversion engine, with structural analysis separated from rendering. Handles paragraph reconstruction, list detection, heading inference, and conservative table extraction with fully explainable outputs. No OCR, no ML, fully deterministic. Built as the foundation for a document ingestion pipeline: clean, deterministic parsing is the layer that makes downstream ML reliable.

Evolving into a full document processing suite with multi-format conversion and PDF compression.

• Python
• Document Processing
• Systems Design