Hi there! I'm Caleb

🌱 Currently diving into Multi-paradigm Design in C++

📫 Best way to reach me: Email preferred. Cell available by request for faster back-and-forth. NashDev Slack

📍 Nashville, TN

I'm a software engineer passionate about low-level systems, AI-driven tools, and cross-platform libraries. Lately, I've been building C#/.NET components for voice processing, speaker diarization, and intelligent agents, alongside some model fine-tuning. Check out my recent work below!

Spotlight Project: MinimalTextClassifier

I'm excited to highlight my latest model fine-tuning work: MinimalTextClassifier!

This is a minimal .NET ONNX-based binary text classifier with Python fine-tuning support, powered by the Deberta transformer model. It's designed for efficient text classification tasks, bridging Python for model training/fine-tuning and .NET for lightweight, cross-platform inference via ONNX Runtime.

Key Features:

• Fine-Tuning Support: Use Python scripts to fine-tune Deberta on your custom datasets for binary classification (e.g., sentiment analysis, spam detection).
• ONNX Integration: Export models to ONNX for seamless deployment in .NET environments—fast, portable, and no heavy dependencies.
• Minimalist Design: Focuses on simplicity and performance, making it easy to integrate into apps or pipelines without bloat.
• Cross-Platform: Runs on Windows, Linux, macOS via .NET.
• Example Use Cases: Custom text classifiers for smart home devices (Alina, what time is it?), content moderation, or data processing tools.

If you're into AI model optimization or need a quick way to deploy fine-tuned transformers in production, check it out! Contributions welcome—let's collaborate on expanding it to multi-class or more advanced scenarios.

Other Featured Projects

Voice AI & Telephony Tools (Recent C# Focus)

These are my latest C# repos, emphasizing minimal, efficient libraries for speech detection, diarization, and agentic workflows. They're battle-tested in noisy environments and utilize AI models like Silero VAD, ecapa-tdnn, whisper, and phi3.

• MinimalSileroVad
  Cross-platform .NET library for VAD (Voice Activity Detection) and speech segmentation using the Silero VAD model. Replaces WebRtcVadSharp—faster and more accurate for real-time apps.

• MinimalVoiceAgent
  A lightweight voice agent framework with tool calling, local TTS/STT, and Grok API integration. Build conversational AI without the bloat—ideal for prototypes or local setups.

• SipBotOpen (Highlighted Project)
  Open-source telephone voice assistant with tool calling and PSTN connectivity. Dial it live at 1-615-899-3280! Uses SIP client in C# .NET for VoIP PBX magic. Currently testing VAD, AEC, and noise suppression.

• MinimalDiarization
  .NET library for AI-powered speaker diarization via ONNX runtime. Includes experimental intent preprocessing for command discrimination in noisy, multi-speaker scenarios—great for voice agents.

Other Highlights

• XMapLib
  Low-level C++ library for mapping Xbox controller inputs to keyboard/mouse. Closer to the metal for gaming/input hacks.

• StreamToActionTranslator
  C++23 CMake project for translating input streams to actionable function calls. Experimental fun with streams and actions.

• youtube_skip_button_fine_tuned_yolo Yolo v11 model fine-tuning scripts for youtube "skip ad" button detection with high-performance inference and excellent accuracy.

More code, tools, and experiments coming soon. Contributions welcome—fork away!

HF: https://huggingface.co/calebt9990