Successes our track record
is our credibility

The Problem

The customer's new book sortation hub had multiple critical software problems, and, after nearly 2 years in development, and with 6 weeks before go-live, it was clear that key components were not going to work well enough at scale.

Our Response

We designed, built, tested and deployed a replacement system for a critical component that drove the sortation machine in 5 weeks, from scratch. The software we wrote was the gateway between SAP, a very large sortation machine, and 200 manual operators.

The Problem

The company had always transformed and reinvented its business to adapt to changing circumstances, and realised in 2015, that as a (mostly) non-digital company, it needed to change to keep up with Industry 4.0. At the time, the company needed a more digitally skilled, engaged and empowered workforce, and to reduce dependence on external vendors, who overcharged and underdelivered, and whom the company could not manage effectively.

Our Response

We proposed (supported by aligned recommendations from McKinsey), and built an internal team which grew to∼100 expert software engineers and developers, to help the company with its digital transformation. We created the right vision, culture (which we subsequently adopted for ourselves), and systems, (independent from traditional IT), and hired the very best people, screening them with a challenging technical test. We used open-source software to radically lower the cost-of-entry and maximise control, flexibility, and quality. The company invested many £m to do this.

The Problem

We identified the need for a warehouse management system which could win new contracts with smaller customers, at lower cost, with fast implementation (where a SAP system would be too expensive). We also needed a system which was flexible and agile, so we could respond to problems and adapt quickly.

Our Response

This drove the development of the DES, based on an open-source platform, and we initially deployed it for a supercar manufacturer. The system became one of the company's two WMS's of choice, and it provides real competitive edge to them, as it continues to develop, and sell as a standalone external solution. Because we built it ourselves, it isn't just lower-cost (no license-fees) but also much better (more flexible, faster to change, and more capable). The end-users in the warehouses became some of our strongest advocates, and best salespeople.

The Problem

We needed the full set of open-source infrastructure to run our own development, test, QA, and production systems. We needed something flexible, adaptable, and low-cost, which could also run the internal and external business systems. It was essential to have full control and auditability of privacy and security for data-protection/GDPR, and to host in a data-centre and location of our choice.

Our Response

We built our own instance of OpenStack (private cloud), which proved to be extremely reliable and cost-effective. OpenStack is "The Most Widely Deployed Open Source Cloud Software in the World", and cost us approximately 1/3 the price of the commercially-hosted alternatives, and making further savings by using open-source for the guest operating systems, platforms and libraries. This is a critical part of the capability needed to develop our systems at speed and deploy quickly.

The Problem

We have been requested to analyse multiple concepts for renewable energy, and to provide an independent assessment of the claims for the proposed designs, and any suggestions for improvements.

• Wave Energy Generator: a proposed design "capable of generating electricity at 2p/kWh", with a claim based on the inventor's experience of resilient marine-engineering on oil platforms.
• Wind Energy Storage: aiming to "arbitrage cheap surplus-energy from wind-turbines at night, reselling it at higher prices during the daytime", with a plan to store the energy as gravitational potential energy in mine-shafts.
• Vortex Energy Recovery: aiming to "recover wind energy, lost by vehicles via aerodynamic drag, converting the vortices back to electricity", where something that is intuitively obviously true turned out to be mathematically and experimentally wrong. 
• Others: we have also analysed a number of other designs, about which we cannot comment further here.

Our Response

We did a detailed analysis of the systems, wrote clear reports, and explained it to the customers. To do this, we needed the physics and mathematics for the mechanical system-dynamics; an understanding of the engineering approaches that were being taken, and the possible alternatives; an understanding of the electrical systems; and to know about the business-constraints, the manufacturing-costs, and the economics. For the analysis, it is most helpful to consider conservation-of-energy, or energy-cycles, and scaling laws.

The Problem

NATS, together with the CAA, BMAA, and LAA were promoting uncertified GPS position-sources for Mode-S transponders, this gave the pilot a low-cost way of transmitting their position out via ADS-B. Thus, allowing the emerging multitude of ADS-B receivers to accurately show the position of nearby aircraft, to increase situational awareness. However, there was no method for the pilots to test their transmission, to ensure their position was being correctly sent, and the correct parameters (SDA, SIL, ...) were set (a requirement by the regulatory bodies). At the time the only affordable way to test this was via a Pilotaware box, however, this showed its latitude and longitude, not that of the aircraft.

Our Response

The solution was to write a software application that utilised the RTL-SDR open source library to drive a low-cost SDR (Software Defined Radio), to pull in the data transmissions and then to decode and display this in an easy-to-use piece of software.

The Problem

Back in 2003/2004, Business travel was, and had long been, dominated by dumb-terminal software booking system, as they were perceived as both faster and requiring great skill to use. Many attempts were made to introduce a rich GUI based booking system, however, these were rejected by the industry as being too slow, thus reducing productivity, and by the operators as they "de-skilled" them.

Our Response

The solution, which still forms the basis today of a global reservation system, was a smart terminal. It allowed the skilled agent to make their quick cryptic entries, while allowing new agents to simply type the first letter to be guided through the booking process, a sort of self-training program. The core system's responses were then made clickable, allowing even greater productivity for the new agent and seasoned professional alike.

The Problem

The bespoke van-sized crash-forming and roll-forming machine formed a critical part of the manufacturer's production line, creating the initial parts from sheet-steel, which it had to die-stamp, then fold, then roll. These parts would subsequently have internal pipework added, and then be welded together by a robotic welder, making about 30 assemblies an hour. However, about 1/3 of the parts were created in a severely distorted shape, leading to subsequent weld failures and leaks.

Our Response

We formed a hypothesis about what might be happening within the internals of the machine, and set up photographic equipment to investigate: this required specialist lenses and lighting to obtain high-quality high-speed video images of the internals, while standing outside the safety-cage, at just the right angle to see into the middle. We then used frame-by-frame motion-analysis to discover exactly what was happening.

The Problem

The customer has a large number of very complex internal business processes (HR, Finance, Procurement, Operations, etc), which involve document-flow through multiple stakeholders, signoffs, and with sometimes very baroque and convoluted rules. This used to be manual, very inaccurate, and time-consuming. Although the off-the-shelf providers for business-management software exist, quite often they require the business to to adapt its processes to the software, instead of the other way around.

Our Response

We investigated various tools, and concluded that the business-logic requirements were so complex and bespoke that it would be better to build a range of extranet applications to cover all the different business processes, such as management of Health and Safety, and Employee Change processes. The solution was a re-useable code platform, based on Django, that could be used to rapidly develop new applications when the need arose.

The Problem

This had become a complex process, needing to track bookings, avoid conflicts, physically book the instruments out-and-back-in, issue invoices, take/check payments, and answer queries. It's also necessary to monitor the repair-status of the instruments, and track the process between the treasurer, instrument-manager, custodian, and hundreds of hirers per year. Percussion instruments are complex, expensive, unwieldy, and confusing. 

Our Response

We decided to automate the old paper process, and designed and built the TIMPANI system out of Linux, Postgres, PHP. This supports the collaboration of the multiple stakeholders, specific-workflows, and specific business-logic: discount policy, insurance rules, membership rules, etc.

The Problem

The paper-based tool had to be simple in order to be fast to use, so it had limited functionality, no long-term data tracking, no data-quality checks or standardisation, and go-and-see required physical access. There was no aggregate reporting, or integration with other systems. Some tools were hard to use and understand, and needed duplicated data-entry (notably PDM).

Our Response

We worked to deeply understand the customer needs, tools, and processes, and then built this using an agile approach, constantly improving and iterating. It began as a shared repository for PDFs, before becoming fully data-driven. We used open-source tools (PostgreSQL, PHP, JQuery, Bootstrap, Plotly etc), and a GPL'd framework, to speed up the development process; nevertheless it is a large and comprehensive codebase (1725 A4 pages-worth of source-code, or a sloccount value of £6m). We hosted it on OpenStack, where it has proved exceptionally reliable. We re-designed the standard paper formats to take advantage of being digital, making them easier to use, with integrated help, richer details, one best way across all users, and making the data-entry "happy-path" fast.

The Problem

The business was creating a tremendous amount of data, that it was not fully leveraging; it also suffered from 'data gaps' due to missing data collection steps in end-to-end processes, and poor quality data-collection which meant that the data we did have, couldn't be properly analysed. Some data was input or transferred manually, leading to mistakes and errors.

Our Response

We led the creation of a suite of tools and processes to increase the amount of useful information we can produce from existing and new data, making use of open-source libraries, in particular, numpy/scipy (numerical-python). These included: data pipelines; exploratory data analysis; forecasting, optimisation and simulation; business intelligence and self-service dashboards; bespoke dashboards; machine learning; and predictive analytics.

The Problem

The customer, a leading UK health & fitness provider needed to improve the user-experience and effectiveness of their website landing-page. They were running pay-per-click advertising campaigns to direct users through to their website; however, they were having difficulty turning those visitors into conversions, and asked us to undertake research on why, and to recommend changes.

Our Response

We recruited participants from the target market, to undertake pre-defined tasks in trials of usability / UX testing sessions, and recorded their on-screen interaction, as well as audio visual data of their participation. This information, coupled with our own insights, helped identify problems, and we explained this to the client. For example, the website didn't communicate the brand's reputation well enough; we overcame this perception by adding information about their 25-year track record, industry leadership, and testimonials. We also discovered, by 'heatmapping', that users were being distracted by the site's complex navigation system, and not reaching the enquiry form; we made it more prominent, and removed the distractions from the landing page. We continued to monitor the pages, making small incremental changes at key points in the campaign, observing their effects using A/B testing methodology.

The Problem

Our customers needed a performance appraisal process that can: add precision to the definition of behaviours; incorporate an assessment of observed behaviours in the appraisal process; aggregate team and organisational performance to learn from general trends; and improve our ability to feedback to improve performance and confidence. Each customer's process and requirements is different.

Our Response

We created a platform to provide a secure, confidentiality-assured, process to enable people to participate fully in a periodic performance-review, embracing both behaviours and results. ToPiC has a very flexible data-structure and analysis framework, allowing both self-reflection, and pseudonymised feedback between team members. It supports many question/answer formats, with embedded help, and has flexible administration tools (to check progress, chase-up, and view/redact answers if need-be). ToPiC creates graphical reports for each user, for a team overview, or for all-participants, showing a customised set of charts, including box/whisker and Ansoff-matrices.

The Problem

The organisation was set up many years ago, and had little technological infrastructure. This meant there was no single point of truth about the international members, no way to update and disseminate know-how (most training was paper-based), and needed a technological strategy and implementation. It was also essential to have a unified brand, to keep long-term control of their own destiny, and to minimise costs.

Our Response

While Telos Partners helped the customer adopt a growth-strategy, Telos Digital identified and recommended a technology-strategy, which they accepted. We built, customised, branded, configured, and hosted the entire set of business-solutions that the customer needed: a learning-management system; a single-point-of-truth and single-signon user database; e-mail; documents, calendar, files; a website and content-management system; knowledge-management system; and built a custom user-interface for admin/management. The technologies include: Moodle, WordPress, Keycloak, NextCloud, MediaWiki, PostgreSQL, …).

The Problem

Customers need to use LLMs in way which is relevant (using their data and know-how), private (keeping their data from leaking into a public training dataset), and at a reasonable cost (hundreds, not millions, of pounds). Building their own foundational model (or training another one) would have been unfeasibly costly, whereas just using a public LLM, unmodified, would mean that the model could not assist with domain-specific queries. A substantial degree of R&D and experimentation was required, together with a detailed understanding of the customer's own problem-domain. Furthermore, the state-of-the-art is changing faster than any other recent technology.

Our Response

We evaluated, tested, researched and integrated multiple technologies, learned how LLMs "think", how they scale, what the performance limits are, where they excel, and how and why they fail. For example, LLMs perform well at memory-tasks, but cannot do cognitive-tasks, unless they have seen the problem before and "remember" the answer. Neural networks are trained in a way that rewards answers that are correctly formatted and well-structured, but not necessarily true, so A.I.-errors are often subtle and look plausible. We found, and constantly review, the technical infrastructure needed for good performance, at a good price. The best current approach is RAG (Retrieval Augmented Generation), using the ChromaDB vector-database, and the Llama 2 (updated: Llama 3) model. This can be run locally, or via inference-as-a-service.  We built, tested, and carefully optimised a tool, L11g, using open-source components, that could interface customer documents to inference-as-a-service, and was easy to use and integrate.