Skip to content
R&D & Exploration

The Experiments Lab

Where curiosity meets code. Real exploratory R&D into local LLMs, embedded IoT, computer vision, spatial computing, and autonomous agents:

Completed
Local AI & LLMs

Offline Local LLMs & Embedded Inference

The Hypothesis / Idea:

Can an organization run zero-data-leakage intelligence models completely offline without sending enterprise data to external cloud APIs?

The Technical Experiment:

Benchmarked quantized Llama 3 & Mistral models running on local Apple Silicon & Linux edge servers with custom vector databases (ChromaDB).

Key Insight / What I Learned:

Local quantized models are now fast enough to replace cloud APIs for 70% of routine corporate document processing tasks.

OllamaLlama 3Mistral 7BPythonChromaDBFastAPI
Active Prototype
IoT & Hardware

ESP32 Environmental & Attendance Sensor Node

The Hypothesis / Idea:

How to automate smart classroom attendance and indoor air quality telemetry without high-cost proprietary hardware?

The Technical Experiment:

Engineered an ultra-low-cost ESP32 microcontroller with RFID reader, temperature/CO2 sensors, and MQTT transmission directly to school ERP.

Key Insight / What I Learned:

Hardware edge computing combined with lightweight MQTT protocols can reduce smart classroom infrastructure costs by over 75%.

ESP32C++ / ArduinoMQTTNode.jsWebSocketsHardware Design
Research Lab
Computer Vision

Real-Time Object & Gesture Detection on Edge

The Hypothesis / Idea:

Can low-cost webcams turn any standard projector screen into an interactive digital smartboard via real-time computer vision?

The Technical Experiment:

Built OpenCV and MediaPipe pipelines in Python to track optical laser pointers and hand gestures, translating coordinates into digital whiteboard strokes.

Key Insight / What I Learned:

Computer vision can democratize smart classroom technology for rural schools with zero requirement for costly capacitive touch panels.

OpenCVMediaPipePythonNumPyCanvas API
Completed
AR / VR & Spatial

Interactive WebXR Physics Playground

The Hypothesis / Idea:

How to make complex atomic orbitals and gravitational physics accessible directly inside a web browser without installing 5GB apps?

The Technical Experiment:

Developed interactive WebXR 3D simulations utilizing Three.js and Cannon.js physics engines with full mobile and VR headset support.

Key Insight / What I Learned:

Spatial interactivity dramatically reduces cognitive load when learning 3D geometric and physics concepts compared to static 2D diagrams.

Three.jsWebXRJavaScriptGLSL ShadersCannon.js
Live Experiment
Agentic Automation

Self-Healing Workflow & Research Agent

The Hypothesis / Idea:

Can an autonomous AI agent monitor regulatory filings, summarize changes, and alert compliance teams without human initiation?

The Technical Experiment:

Created an asynchronous LangChain/LangGraph agent pipeline that scrapes regulatory updates, validates changes with structured LLM schemas, and dispatches structured digests.

Key Insight / What I Learned:

Deterministic guardrails around probabilistic LLM agents are essential for enterprise compliance workflows.

LangGraphPythonPlaywrightOpenAI / Claude APIsCron Tasks