theLAB
Learning. Algorithms. Breakthroughs.
Practical AI research, local-first systems, and home-lab infrastructure documented from experiments that actually run.
Deep Dive Into AI
Master the architectures, algorithms, and methodologies powering the next generation of intelligent systems.
Every topic is framed around buildable systems: what to study, what to run locally, and where the tradeoffs show up when research leaves the notebook.
Neural Networks
Build and train deep neural networks from scratch. Understand backpropagation, activation functions, and optimization.
Transformers & LLMs
Explore attention mechanisms, transformer architectures, and large language model fine-tuning with Ollama.
Computer Vision
Image classification, object detection, segmentation, and generative models for visual understanding.
Reinforcement Learning
Train agents to make decisions through reward optimization. Q-learning, policy gradients, and multi-agent systems.
AI Security
Adversarial attacks, model robustness, prompt injection, and secure deployment of AI systems.
Production MLOps
Deploy, monitor, and scale AI models. Docker, Kubernetes, model serving, and continuous integration for ML.
Hey ADI — Real-Time Vision
Realtime mode for ADI — continuous vision, conversational pace, no tool calls, no "send message" button. Just a live conversation with something that's actually watching. Tracks intent through speech, recognizes detailed objects, and reads OCR-grade text. All running on theLAB hardware.
Things We've Built
Self-hosted AI experiments and home-lab infrastructure. Production systems running on the LAMP + Ollama stack.
ADI — Advanced Data Intelligence
Open-source dashboard that unifies your AI fleet. Local Ollama hosts plus cloud APIs (Anthropic, OpenAI, Google) behind one streaming chat surface — with voice in/out, camera vision, gesture control, persistent memory, and a built-in MCP tool layer.
ADI Montage
Self-hosted agentic video studio. Type an idea and an agent researches, scripts, generates imagery and motion, narrates, scores, captions, and renders a finished MP4. Local Devstral drives it; FLUX/ComfyUI, Kling, ElevenLabs, and Remotion do the work — cost-aware, with a TikTok 9:16 mode.
ADI Distilled Models
A family of small, fully local models distilled from frontier teachers and served from Ollama. Organized by line — one per student base (Qwen3, Gemma, Llama) — with each model distilled from a different teacher for a different purpose. First live model: a 2.7 GB Qwen3.5-4B distilled from glm-5.2.
AI DJ Radio
24/7 self-hosted radio station. Local LLM writes DJ scripts, Qwen3-TTS voices two personas rotating by daypart, Liquidsoap assembles the stream, Icecast broadcasts. Auto-generated news breaks from RSS + NewsAPI, plus weather breaks pulled live from the NWS API.
AI Vulnerability Scanner
AI-powered vulnerability scanner that combines automated reconnaissance with local LLM analysis. Identifies security weaknesses, classifies CVEs, and generates remediation reports with contextual threat intelligence.
Custom Wakeword Neural Network
Custom openWakeWord model trained from scratch to trigger ADI Genesis on "hey addie." Built from 8,300+ synthetic positives across 41 Kokoro TTS voices, augmented against ACAV100M backgrounds, and exported to ONNX.
Ready to Push Boundaries?
Join theLAB community and be part of the next breakthrough in artificial intelligence research and development.