dida

BACKGROUND

Please describe your company and your position there.

I am the CEO of valuedriven, an innovative web-development agency, leveraging a network of vetted partners in order to solve interesting problems with a wide range of technologies. We are working together with companies to create a detailed step-by-step technology plan towards success.

We do strategic project planning and software development to increase the value that companies get out of their technology stack. We focus on web and mobile implementation projects with a wide range of technologies like React, WordPress, Angular, NodeJS, Django/Python and choose the best tools for the use case at hand.

OPPORTUNITY / CHALLENGE

For what projects/services did your company hire dida Machine Learning, and what were your goals?

valuedriven was hired to create a state-of-the-art computer vision project for a large client from Germany in order to speed up the a highly manual and complex process. We developed the front end of the solution but the backend involved complex pattern recognition and advanced algorithmic modelling.

For this we needed to turn to a machine learning agency and dida was our choice. The challenge was to develop a solution that needed none or only little human involvement to automate the process which had previously relied heavily on manual work. Also, certain maximum computational reaction times had to be met.

SOLUTION

How did you select this vendor and what were the deciding factors?

We were looking for a software solution provider that really excels at algorithmic problem solving and out of the box thinking. dida had worked on complex pattern recognition projects and seemed to have the expertise for the problem at hand.

Describe the scope of work in detail, including the project steps, key deliverables, and technologies used.

dida developed the solution within an agile project framework (first MVP, continuous user feedback, eventually production-ready software), achieving an automation of the entire process. It was a very pleasant cooperation, communication went very smoothly and deadlines were met.

The production pipeline involved different modules – amongst them an astonishingly precise image segmentation by means of deep neural networks. The whole functional solution was shipped as a docker container and could be implemented on AWS within days. It has been running stable ever since.

How many people from the vendor's team worked with you, and what were their positions?

One Solution Architect and 3 Machine Learning Scientists.

RESULTS & FEEDBACK

Can you share any measurable outcomes of the project or general feedback about the deliverables?

dida’s expertise in computer vision and machine learning was critical to building and fine-tuning the automation engine. In addition, their focus on sound project management and good proactive communication was instrumental to delivering on time and within budget.

Describe their project management style, including communication tools and timeliness.

We used Slack for communication and they were always very proactive.

What did you find most impressive or unique about this company?

We were very impressed with their deep technical knowledge and their ability to explain it in plain terms, which is very crucial especially when we are talking about such a complex domain like neural networks.

Are there any areas for improvement or something they could have done differently?

We always felt that we were in good hands and that the people involved were experts in their field. I can’t think of anything. The project ran smoothly.

BACKGROUND

Please describe your company and your position there.

I am the CEO of publicplan, an innovative IT company specializing in the digitization of German public administration services. We develop web- and open-source-based software solutions for public sector institutions. When it comes to e-government consulting, we focus on technology-related topics where we can optimally incorporate our experience in software development into the IT consulting process.

OPPORTUNITY / CHALLENGE

For what projects/services did your company hire dida Machine Learning, and what were your goals?

publicplan developed an online service portal for the federal state of North Rhine-Westphalia which enables citizens to access public administration services. One key feature we wanted to offer was an intuitive search function for industry codes, integrated both in the portal's chatbot and the business registration assistant. Citizens registering a new business in Germany have to provide an industry code along with their registration. This industry code is chosen from a list of over 800 different codes, each described and defined in complicated “public administration language”. Finding the correct code from all these options is hard and the input to our algorithm needed to just be a free-text description of the business entered by the business owner. An added challenge was that official business descriptions that accompany the codes are very different from everyday language usually used in the free text.

SOLUTION

How did you select this vendor and what were the deciding factors?

We found the vendor through our personal network (recommendation).

Describe the scope of work in detail, including the project steps, key deliverables, and technologies used.

The solution developed by dida’s team involved adapting and training an algorithm based on state-of-the-art AI architecture developed to solve a variety of Natural Language Processing problems. The core model was pre-trained on a large German language corpus and was then enhanced and fine-tuned by dida to address our specific problem, making sure it is preformant on a CPU if a GPU is not available.

This final classifier was trained using historical business description - industry code pairs. The 5 highest-ranking industry codes are then sent back as a response to the original user request in ranked order. The solution was both flexible and easy to maintain. New industry codes can be integrated with very little effort and the deployment can connect to various API endpoints. We ended up integrating it with both our chatbot and our business registration assistant.

How many people from the vendor's team worked with you, and what were their positions?

In total we had contact to 3 people from dida.

RESULTS & FEEDBACK

Can you share any measurable outcomes of the project or general feedback about the deliverables?

Prior to working with dida, we had implemented a solution using basic word embeddings, which often showed unsatisfactory results. This indicated that better semantic understanding of the definitions and descriptions of different industry codes, as well as using more colloquial descriptions, were needed. The outstanding performance of dida’s model is a testament to how well the team utilized semantic understanding to solve the problem.

Describe their project management style, including communication tools and timeliness.

Communication with the team at dida was always top notch and we felt that the Machine Learning Scientists were deeply engaged with the project from conception to integration with our existing platform. The accessibility of the team and their willingness to explain their thought process in detail created an atmosphere of collaboration and comfort with the decision-making.

What did you find most impressive or unique about this company?

Finally, their NLP knowledge and expertise are impressive and it really made a difference that they are very involved with the research side of the field.

Are there any areas for improvement or something they could have done differently?

Currently none.

BACKGROUND

Please describe your company and your position there.

I am Director Product for felmo’s medical services. We provide at-home visits by highly rated veterinarians for a stress-free experience, plus a wide variety of digital features rounds out the holistic pet health service. We currently have 200+ employees and are based in Berlin.

OPPORTUNITY / CHALLENGE

For what projects/services did your company hire dida Machine Learning, and what were your goals?

Our customers provide free-text descriptions of the reasons why they need a vet appointment for their pets. We wanted to automate the assessment and analysis of these customer requests to generate the metadata in order to improve our appointment preparation and scheduling.

SOLUTION

How did you select this vendor and what were the deciding factors?

We were looking for a machine learning company with a proven track record of executing information extraction projects.

Describe the scope of work in detail, including the project steps, key deliverables, and technologies used.

The scope of dida’s services ranged from assessing potential Machine Learning use cases to strategic advice on how to proceed in the area of ML after the end of the project. This included:

• Assessing our current internal challenges for their potential ML optimisations and support in selecting the field where most business value generation was expected
• Formulation of the selected challenge as an ML problem and definition of performance metrics
• Defining a data labelling scheme and support our internal field experts during the process of data annotation
• Ensure high data quality in close communication with our field experts
• Training and evaluation of machine learning models with different complexity
• Advise on best options to integrate the solution into our existing workflows
• Advise our internal data science team on best practices for deployment

How many people from the vendor's team worked with you, and what were their positions?

• On our side, the Director Product and a Data Scientist were responsible for the project.
• On dida’s side, three machine learning scientists were working on the project.

RESULTS & FEEDBACK

Can you share any measurable outcomes of the project or general feedback about the deliverables?

The project was successful, the developed solution allows us to extract structured data from our customers requests with high performance. dida explained the pros and cons of different approaches in detail and also analysed potential areas of even further improvement in the future. dida also advised us on how exactly the solution can be used in production and how to configure it, depending on our changing requirements, in a comprehensible way.

Describe their project management style, including communication tools and timeliness.

Project management and communication were straightforward with short, effective communication cycles on a high level of shared understanding. We had regular meetings to discuss project progress where the current status was clearly communicated and ML related decisions were well explained.

Upcoming tasks were resolved in a very pragmatic and time efficient way. We appreciated that they understood our fast paced start-up environment and the challenges inherent to our specific area of business.

What did you find most impressive or unique about this company?

Their domain expertise in different machine learning processing models was elemental to the success of our project. Also, dida has taken the time to deeply explain and guide us through different technological tradeoffs, giving recommendations, which we could use for another iteration, making their model more accurate for our purposes.

Are there any areas for improvement or something they could have done differently?

-

BACKGROUND

Please describe your company and your position there.

I am a Technology Strategy Manager at Accenture Strategy and Consulting. We provide business strategy, technology strategy and operations strategy services as well as consulting services. Accenture works for more than three-quarters of the Fortune Global 500. We are one of the biggest consulting firms in the world with over half a million employees.

OPPORTUNITY / CHALLENGE

For what projects/services did your company hire dida Machine Learning, and what were your goals?

We always need to be up to date with current technological developments to provide the best value to our clients. Therefore, we were looking for a partner to offer workshops on AI and machine learning for our consultants

SOLUTION

How did you select this vendor and what were the deciding factors?

We found them through a referral and compared their offering as well as their pricing to other vendors in this area. The offering convinced us and the pricing was in "the usual" price range.

Describe the scope of work in detail, including the project steps, key deliverables, and technologies used.

The scope of dida’s services included the preparation of three different courses:

• AI and deep learning
• Deep neural nets powering modern computer vision
• State of the art language models in NLP Each course had:

1. a duration of half or full day
2. theoretical as well as practical “hands-on” parts
3. a maximum of 5 participants

How many people from the vendor's team worked with you, and what were their positions?

On dida’s side a Machine Learning Scientist and a Machine Learning Consultant gave the workshops. I also had contact to the company's CEO (referred contact) and a back-office colleague dealing with PO, invoicing, etc.

RESULTS & FEEDBACK

Can you share any measurable outcomes of the project or general feedback about the deliverables?

In our pilot series of 4 workshops, a total of 17 employees attended the training. All attended employees recommended the training sessions to other employees.

Describe their project management style, including communication tools and timeliness.

The cooperation went very smoothly, there was a single point of contact and dida was always ready to answer our questions and adapt their courses.

What did you find most impressive or unique about this company?

This company manages to transfer the Berlin Start-up character to our big conservative company while being perfectly professional.

Are there any areas for improvement or something they could have done differently?

The team has been great and we’re pleased with the professional organization and communication.

BACKGROUND

Please describe your company and your position there.

I am a Senior Manager in Good Governance and Responsible Mining at Levin Sources, a consultancy and advisory firm specialised in responsible mining and mineral value chains, in particular focussing on artisanal- and small scale mining (ASM). Working with actors in the public, non-profit and private sectors at all stages of the mineral value chain from upstream to downstream.

We have a global and thorough understanding of the challenges and opportunities in promoting responsible mineral value chains. We offer services in research, strategy, policy, capacity building, and due diligence, and have worked on projects that have generated impacts in more than 50 countries globally. We currently have 15 employees and are based in Cambridge, United Kingdom.

OPPORTUNITY / CHALLENGE

For what projects/services did your company hire dida Machine Learning, and what were your goals?

Jointly with dida, we wanted to automate the monitoring and analysis of artisanal small scale mining sites in the Amazonian rainforest and other areas. Through the use of machine learning, large and inaccessible regions can be monitored and thereby ASM baselining and change detection becomes cheaper, more efficient and more viable.

SOLUTION

How did you select this vendor and what were the deciding factors?

dida came to our attention as a technology partner supporting us with the development of an automated tool to scale and support our monitoring activities.

Describe the scope of work in detail, including the project steps, key deliverables, and technologies used.

The scope of dida’s services include:

• Acquisition of remote sensing satellite imagery
• Setup of AI pipelines and infrastructure
• Setup and deployment of a web interface to use the tool
• Preparation of datasets for labelling by us
• Technology consulting

The project started in summer 2020 and is still ongoing. The project budget lies within the 25k - 50k€ range.

How many people from the vendor's team worked with you, and what were their positions?

On our side, one Senior Manager and one Senior Specialist are involved in the project.

RESULTS & FEEDBACK

Can you share any measurable outcomes of the project or general feedback about the deliverables?

The project is running very successfully. New requirements like new satellite technology (Radar) are included on request. Technical project proposals for our clients are written quickly and with confidence. The dida team is always at hand to present and explain the technology solution to our clients.

Describe their project management style, including communication tools and timeliness.

Project management and our collaboration goes very smoothly.

What did you find most impressive or unique about this company?

They are always ready to answer our questions and adapt their work to our ever-changing business context. A good way to check if an idea to apply machine learning makes sense also from a business perspective is to just talk to them in advance and let them look at sample data, as this comes at no additional cost.

Are there any areas for improvement or something they could have done differently?

Nothing to report. We can fully recommend working with dida.

BACKGROUND

Please describe your company and your position there.

I am the CEO of a German online tenant protection portal. We connect our customers with selected partner attorneys who provide legal advice. Our portal’s members can ask their questions online and upload necessary documents (such as rental agreement or utility bill). Then our partner attorneys will provide an assessment within 48 hours. We currently have 28 employees and are based in Berlin.

OPPORTUNITY / CHALLENGE

For what projects/services did your company hire dida Machine Learning?

We wanted to automate the assessment and analysis of documents typically uploaded by our customers (rental agreements and utility bills) as far as possible - not to replace, but to to assist our partner lawyers. The measure of success would be the factor by which we could reduce the time our partner attorneys needed to accomplish the respective legal checks.

We expected this project to require the use of machine learning methods and expertise in the particular areas of natural language processing and information extraction - which we did not have in-house.

SOLUTION

How did you select this vendor and what were the deciding factors?

dida came to our attention because we were looking for a German machine learning company (because the project would require native speaker level German) with experience in NLP and information extraction projects. dida then convinced us with a prototype to entrust them with the project.

Describe the project in detail and walk through their service package.

The scope of dida’s services ranged from preprocessing and labeling raw data (like document scans) to assistance with the deployment of the created software. This included:

• Acquiring legal expertise to define sound labeling schemes
• Definition of tasks and performance metrics
• Training and evaluation of various task-specific machine learning models
• Making predictions and results explainable
• User testing workshops with partner lawyers
• Delivery of dockerized app with customized API

How many resources from the vendor's team worked with you, and what were their positions?

On our side, one product owner was responsible for the project, for the deployment supported by our in-house IT. dida had three machine learning scientists working on the project. For the project we followed the principles of agile software development, although not in an orthdox manner.

RESULTS & FEEDBACK

Can you share any outcomes from the project that demonstrate progress or success?

The project was a complete success. With the software created by dida, the processing time of customer requests could be reduced by up to 80%. dida took care to present the way the algorithms arrive at their results in a comprehensible and transparent way.

The lawyers working with the software know why they decide the way they do, which makes them trust and use the software. Apart from the successful completion of the project, we were impressed with dida’s technical and theoretical ML expertise. They did not just compute some numbers, but could explain and justify what they did at every step.

How effective was the workflow between your team and theirs?

The project management and the cooperation in general went very smoothly. dida was always ready to answer our questions and adapt their work to additional requirements we came up with. They communicated their progress and results transparently so that we always felt well-informed and up to date. We especially appreciated that they understood the business aspects of the projects very well and always considered them in their decisions, regarding both the overall structure of the project and detailed modeling decisions.

What did you find most impressive or unique about this company?

Simply the high quality of each of their employees and the commitment to get things done in time.

Are there any areas for improvement or something they could have done differently?

We learned that it is important to think about (and communicate) limitations that come with the available server infrastructure and envisaged use cases as early as possible in the project (like acceptable computation times).