Work

SiParadigm

SiParadigm · AI Developer II

Document Digitization Pipeline

Turning scanned and digital PDFs into a structured hierarchy of sections, key-value pairs, checkboxes and tables using vision-language models and OCR.

PythonVision-Language ModelsOCRYOLOasyncioPydantic

SiParadigm · AI Developer II

Medical Document Redaction

Policy-driven redaction of sensitive information in scanned documents, grounded in OCR geometry, cutting manual effort by about 80%.

PythonLLMsOCRYOLOFuzzy Matchingasyncio

SiParadigm · AI Developer II

Configurable Document Data Extraction

Extracting a client-defined set of fields from scanned forms, each with a value, a verified location on the page and a confidence score.

PythonLLMsVision-Language ModelsOCRFuzzy Matchingasyncio

SiParadigm · AI Developer II

Microscope Autofocus: Foreground-Aware Sharpness Scoring

The focus-evaluation stage of an automated microscope's autofocus: score how sharp each frame is, ignoring background and noise, and pick the sharpest height.

C++OpenCVComputer Vision

SiParadigm · AI Developer II

Density-Guided Field Sampling for Slide Scanning

A probabilistic sampler that turns a few marked regions on a low-magnification overview into a ranked list of high-magnification fields most likely to contain cells.

C++Kernel Density EstimationProbabilistic Sampling
Focus Maps with Delaunay Triangulation

SiParadigm · AI Developer II

Focus Maps with Delaunay Triangulation

Interpolating the in-focus height across a microscope slide from a few measurements, using Delaunay triangulation and planar interpolation.

C++PythonPygameComputational Geometry
Laplacian Pyramid Focus Stacking

SiParadigm · AI Developer II

Laplacian Pyramid Focus Stacking

Fusing a stack of microscope images taken at different focus heights into one all-in-focus image, prototyped in Python and ported to C++.

PythonC++OpenCVNumPyImage Fusion

SiParadigm · AI Developer II

Healthcare RAG Pipelines

Healthcare-focused AI pipelines, including Retrieval-Augmented Generation (RAG) systems.

RAG

Freelance

Freelance client (US) · Freelancer

AI Voice Co-Host for Twitch Streams

A real-time AI co-host that reads Twitch chat, listens to the streamer's microphone, and answers out loud in a character voice with low latency.

PythonLangChainGeminiElevenLabsTwitch APIStreamlitFFmpegMultithreading

Freelance client · Freelancer

Rocket League Agent Trained with Reinforcement Learning

A Rocket League bot that learns from scratch through PPO self-play against the live game, then ships as an RLBot-compatible bot.

PythonPyTorchStable-Baselines3PPORLGymRLBotTensorBoard

Freelance client · Freelancer

Smart-Plug Energy Monitoring Service

A multi-user web service that tracks per-appliance electricity use from SwitchBot smart plugs, with scheduled polling, a JWT-secured API and charts.

PythonFlaskCeleryRedisMongoDBJWTSwitchBot APIMatplotlib
Procedural Physics Video Generator with Collision-Driven Sound

Freelance client · Freelancer

Procedural Physics Video Generator with Collision-Driven Sound

Generates short-form 'satisfying physics' videos of nested bouncing circles, with a tone synthesised for every collision.

PythonPygameNumPyOpenCVSciPyMoviePyAudio Synthesis
Monocular 3D Position and Orientation from Coloured Markers

Freelance client · Freelancer

Monocular 3D Position and Orientation from Coloured Markers

Estimating a rod's 3D position, roll and yaw from a single wide-angle camera using colour segmentation and known object size, with no depth sensor.

PythonOpenCVNumPyComputer Vision
Single-Camera 6-DoF Stylus Tracking with QR Markers

Freelance client · Freelancer

Single-Camera 6-DoF Stylus Tracking with QR Markers

Tracking a hand-held stylus relative to a test surface in 3D with one calibrated webcam, two QR markers and Perspective-n-Point.

PythonOpenCVNumPySciPyCamera CalibrationPnP

Freelance client (UK) · Freelancer

Single-Camera Robot Localization

An application that localizes robots within an arena using visual markers and a single camera.

Computer Vision

National Center for Artificial Intelligence

National Center for Artificial Intelligence · Research Associate

Ad Recognition & Program Classification

Recognising which ad is airing on live TV in real time: CNN frame embeddings matched against an ad database, logo detection with YOLO, and scene detection for timing.

CNNEmbeddingsCosine SimilarityYOLOScene DetectionONNXGPU Inference

National Center for Artificial Intelligence · Research Associate

Large-Scale Social Media Sentiment Analysis

Large-scale sentiment analysis on social media data.

NLP

National Center for Artificial Intelligence · Research Associate

Urdu Speech Recognition for Broadcast Media

An Urdu automatic speech recognition (ASR) system for broadcast media.

ASR