Projects
SightBit
SaaS (TRL-9)Public
- Advanced **computer vision** platform delivering real-time **drowning predictions**, **human and vessel detection**, and **flood-risk alerts**.
- Utilises standard **CCTV feeds** — no specialised sensors or site-specific tuning required.
- Our proprietary **panoptic segmentation model** employs an augmentation pipeline that enhances small-object recall by 18 percentage points.
AdaptiveBridge
Production-ready (TRL-8)Public • Open-Source
- Robust **Python library** for adaptive modelling, designed to predict and fill in missing features during inference, ensuring **ML models** remain resilient in real-world applications.
- Features configurable thresholds for **feature importance**, correlation, and accuracy; supports custom accuracy logic and automated **distribution handling**.
- Provides a user-friendly drop-in estimator: fit once, then call .transform() for seamless integration into existing **pipelines**.
- Efficiently manages feature dependencies, optimising model performance.
- Comes with comprehensive documentation, unit tests, and continuous integration, available on **PyPI** under a permissive license.
VizNet
Initial POCStealth
- Innovative **neural architecture** combining **CNN** and **RNN** with temporal gates for enhanced **video understanding**.
- Tackles issues of class imbalance and motion blur with a camera-aware data curriculum.
- Currently undergoing evaluation on **Kinetics-700** with a baseline improvement of 6 percentage points in **top-1 accuracy**.
MazeByte
SOTA Proof of Concept (PoC)Semi-Stealth
- Showcasing an autonomous **ETL** AI agent capable of self-maintenance for efficient data automation.
- Utilises infant-like feedback loops (reward/pain) and resource-seeking behaviours to foster adaptive learning.
- Generates pseudo-code modules for memory, search, optimisation, and complete **end-to-end ETL automation**.
- Aims to eradicate manual data engineering by automating the entire data pipeline using the proprietary **MBI model**.
Jhive
Production (TRL-7)Public • AIforGood
- A **Django-based web application** leveraging the **OpenAI API** to analyse and detect antisemitic content.
- Offers real-time content classification paired with a detailed reporting interface.
- Provides **RESTful API** endpoints for easy integration while implementing rate limiting.
- Ensures secure input validation and thorough sanitization to prevent injection attacks.