Our Work
Smart Cities Computer Vision
Build for the Indian Smart City projects our engineers developed and deployed a complete set of intelligent video analytics tools to detect car accidents, vandalism and safety issues fully integrated with city CCTV cameras
Read more…
Real Estate Price Modelling
We worked with a leading Proptech company to deliver a price prediction model using not only harvested data but computer vision, LLMs and custom regionalisation algorithms to accurately provide property pricing information.
Read more…
Document Information Capture
Using a combination of text-extraction and LLMs our engineers built a system to extract information for 100s of different types of scanned documents with highly inconsistent formatting. Read more…
Automated Clinic System
Within days of the original covid outbreak our team supported the full process around Pyser Testing and Crown Labs covid testing operations from booking, to back-end IT and provision of results in a fully GDPR and highly controlled privacy environment. Read more…
Traffic Safety
Using a combination of existing CCTV infrastructure and the latest AI video technology our engineers built a system to help protect key infrastructure at one of the busiest port tunnels in Europe. Read more…
Hospitality & Retail
For a number of the world’s leading hotel brands we have built and deployed a variety of custom video solutions, involving both hardware, software and AI technology. Read more…
Smart cities vision
Client: Indian Smart Cities Initiative
Challenge: Indian Smart Cities Initiative had a requirement to automatically identify certain road traffic events in real time, to trigger actions from law enforcement and emergency response teams. These included identifying when motorcyclists were not wearing a helmet, and identifying when two wheeled vehicles were involved in road traffic accidents. Proprietary computer vision systems had proven inadequate when faced with the very busy data received from Indian urban CCTV systems.
Solution: We identified the specific attributes of the data associated with the target events, and built a customised training programme. Fed with specially selected training data, this create a customised computer vision model that was able to identify unsafe riders and with 98% accuracy.
Outcome: The model was taken forward to pilot, and put into action in xx Indian cities, where it was refined further and integrated into the wider Smart Cities system on a regional basis.
Real estate price modelling
Client: Real Estate Consultancy
Challenge: An international real estate consultancy was seeking to automate some of its key processes, including the ability value prospective properties and streamline the appraisal process. Industry-standard AVMs (automated valuation models) are built around the need to value existing assets, and conventional appraisal tools have very little automation and suffer from needless complexity.
Solution: Using a deep learning approach (rather than the programmatic approach more common in the industry) we created a model for valuing prospective properties, and integrated that into a highly automated programmatic appraisal model. The valuation model was 95% accurate across 90% of the client’s target territories, which allowed the client to do away with most of the arduous manual research and guesswork normally associated with this stage of the process.
Outcome: The new model was integrated into a more accurate, more highly automated appraisal model, and between the two innovations, the client was able to pivot their business towards a pure investment consultancy model. This was achieved without the need to hire a team of expensive experienced consultants, as would have been necessary for other organisations without the same technology advantages.
Document information capture
Client: Start-up
Challenge: An innovative start up was trying to create a system to digitise household documentation such as bills and insurance policies. With hundreds of different providers each using a different format, and without access to every possible example, this was a highly challenging AI task. Conventional ‘brute force’ approaches to training an AI model in this scenario would have beyond our client’s budget.
Solution: Our innovative approach was to train an LLM to understand all the different permutations of each term in each kind of document. Using a text extraction algorithm we were then able to standardise data across each document type, and process this into a common data structure.
Outcome: Once integrated into a mobile application, the prototype was able to operate consistently with 99% accuracy across 95% of documents shown to it. Having solved a key technical challenge, the start up was able to move to the next stage of their growth plan.
Automatic clinic system
Client: Specialist Healthcare Provider
Challenge: In the spring of 2020, the covid crisis put unprecedented pressure on the healthcare infrastructures of countries throughout the world. In the UK, the government’s plan to protect the NHS hinged on effective, efficient testing, and experienced organisations were asked to put in place facilities to allow this. Our client was asked to provide a facility in central London.
Solution: Together with the client we created a technology roadmap that could be executed in the minimum possible time. We created a booking portal, database, and results management system, and were able to get these live within four weeks.
Outcome: The new testing facility was able to open on schedule, and to play its part in fighting the covid outbreak in the UK, and protecting the NHS. The booking portal and results management system worked flawlessly, and local workers and residents were able to access accurate testing on the same day.
Traffic safety
Client: A leading outsourced infrastructure management provider
Challenge: A leading outsourced services operator was tasked with managing a road tunnel leading from one of Europe’s busiest port facilities into the city centre. This included providing first-response fire, safety and evacuation services. To make sure the correct resources could be deployed, the operator needed real-time information on the nature of any hazardous substances in the tunnel at any one time. This could be achieved by using CCTV to read the mandatory Hazchem labelling on the backs of goods vehicles.
Solution: We worked with the operator’s technology team to identify a 3 rd party computer-vision solution designed for this specific use case and tuned it for the specific environmental conditions. We then built a custom data management system and user-friendly dashboard so that data could be easily understood and actioned immediately in case of an emergency.
Outcome: The system beat its SLAs and was able to pass stringent tests for accuracy required of a safety-critical applications, including a minimum of false positives, false negatives and mis-reads. The dashboard allows first responders to act correctly whenever there is a road traffic accident in the tunnel. The system continues to keep travellers and workers safe in the tunnel today.
Hospitality
Client: Luxury Hotel Chain
Challenge: A luxury hotel brand was building a new set of super-premium suites, and required an access solution with the maximum of both security and convenience. A biometric recognition solution was identified as offering the best solution, but existing proprietary solutions still required too much interaction from the user.
Solution: Starting with a common biometric recognition model, we fine tuned this so as to minimise the interaction required from the user, and embedded this model in a specific custom machine which could be built into the structure of the remodelled suites (and therefore could not be tampered with by malign actors). We then worked with an electrical engineering company to integrate this with the hotel’s electrical system.
Outcome: Once installed, the system provided complete security for the VIP guests of the premium suites, without entailing any inconvenience. The unique installation ensured that guests could be 100% confident in their privacy and security during their Stay.
Retail video management
Client: National Retail Chain
Challenge: A national retailer with upwards of 50 stores was seeking a way to manage the output from the hundreds of CCTV cameras across its estate. Although proprietary solutions were available, the company’s in house IT team had asked for support in selecting the correct solution and integrating it with the existing systems.
Solution: We were able to identify a proprietary solution, which we could adapt to the specific requirements of the retailers business, and quickly and economically integrate it into the existing hardware and software systems.
Outcome: The retailer was able to effectively manage the output from hundreds of cameras across a nationwide estate, taking care of privacy obligations while quickly and easily accessing specific video clips for purposes including security, safety and customer service.
Client: Casino
Challenge: A casino wanted to explore the possibilities for detecting when gamblers were attempting to cheat. Although casinos routinely employ specialists to look for and investigate this kind of behaviour, the possibilities for using technology to augment and enhance these processes were relatively unexplored.
Solution: We designed a prototype model that was able to combine analysis of CCTV data and win/loss records, and was able to detect when gamblers were attempting to cheat. The model performed extremely well in tests against human specialists.
Outcome: The casino was able to use the ideas outlined in the prototype to move forward in its process for technology development, incorporating it into a broader roadmap covering safety and security across the organisation’s operations.
Gambling cheating detection
Defect detection
Client: Manufacturer
Challenge: Although defect reduction is one of the key goals for all high-tech or high-accuracy manufacturing businesses, proprietary computer vision solutions are not effective for all use cases in this sector. A leading manufacturer wanted to explore the possibilities for increasing the accuracy of defect detection in parts production.
Solution: We were able to create a custom trainable model that could be quickly introduced to the output of any manufacturing process stage, and learn the specific parameters and tolerances that indicate an acceptable part.
Outcome: The model was shown to have an accuracy far beyond that of the available proprietary solutions in the specific use case. The manufacturer was able to use our prototype and take that as a model for a broader, enterprise level solution, delivered as part of a wider technology roadmap.
Retail footfall analysis
Client: National retailer
Challenge: A national retail chain with upwards of 200 stores nationwide wanted to understand the movement of the customers within their stores, in order to optimise the layout of goods and points of sale within the layout. As the stores did not have a common footprint, this required a specific set of outputs for each individual store.
Solution: We built a custom video analysis solution to track the physical customer journey, and understand the relationship of purchasing decisions to product and point of sale placement within a store. This data was then converted into intuitive data visualisations to illustrate the specific actions that needed to be taken in each store.
Outcome: The retail chain were able to optimise the layout of each of their stores, to maximise the number of browser to purchase conversions, and increase the value of each transaction. This had a sustained positive impact on the sales for the business.