Intelliarts

ABOUT THE COMPANY

Our partner is a top electric vehicle (EV) charging management software solution provider in the US market. Intelliarts has collaborated with them for the last few years, being in charge of the development of EV charging management solutions.

BUSINESS CHALLENGE

The company wanted to mitigate this risk of overloading the grid and, hence, add to their value proposition. They contacted Intelliarts to implement the Open Automated Demand Response (OpenADR), the idea of which is to manage the ever-growing demand for charging at peak hours.

SOLUTION

By joining efforts with the customer, Intelliarts implemented the OpenADR 2.0b specification that serves as a gold standard for smart grid operators and sends data and DR signals to shut down electrical devices, such as EV stations, in periods of high demand. As part of the project, we also added notifications about upcoming DR events for site hosts and EV drivers, as well as reporting for utility companies.

BUSINESS OUTCOME

The OpenADR implementation was a big step forward for the EV company, bringing lots of value to our partner, site hosts, EV drivers, utilities and energy service providers. The project helped the company and its customers reduce energy consumption at times of high demand, save costs, and improve grid reliability.

TECHNOLOGIES USED

Java, Spring Boot, MongoDB, AWS SNS/SQS, React

ABOUT THE COMPANY

Our partner is a cybersecurity company that transforms network and cloud activity into evidence and, thus, helps other companies with decision-making.

BUSINESS CHALLENGE

The company reached out to Intelliarts to consult on analyzing large amounts of network security logs to strengthen the company’s insider threat detection and response. One of the core requirements was for the solution to operate within the company’s on-premise infrastructure.

SOLUTION

Intelliarts provided the company with technology consulting services on the topic of integrating large language models (LLMs) locally. In the project, our specialists:

?Thoroughly explained the principles and hardware requirements of LLMs

?Created a roadmap for implementing a hybrid CPU (central processing unit) / GPU (graphics processing unit) approach, adding a dedicated GPU to accelerate LLM inference and optimize performance

?Selected an appropriate pre-trained LLM for summarization and investigated optimization techniques for better efficiency

?Designed a streamlined workflow integrating CPU-based data preprocessing with GPU-powered LLM text generation

?Established a plan of how to integrate the solution into Corelight’s existing systems to support real-time or batch analysis of network logs

BUSINESS OUTCOME

We provided the company with a clear roadmap and technology strategy to significantly reduce the time needed to process and summarize network logs. The insights received can empower our partner to make informed decisions about hardware upgrades and software optimization for faster threat identification. The solution also focused on a locally operated model to consider the company’s commitment to data security and compliance.

ABOUT THE COMPANY

Our partner was a non-governmental organization that advocates for gun control and against gun violence.

BUSINESS CHALLENGE

The NGO has an extensive knowledge base about gun safety and gun violence, and as this base is growing dynamically, the customer finds it more challenging to browse through tons of data and search for the necessary information.

SOLUTION

The Intelliarts ML team built a chatbot using the GPT-4 model, which extracts data quickly through free text requests and interprets it in a manner expected by a user. It also provides references as additional information in case a user wants to verify whether the tool works properly and go deeper into the topic.

When developing the AI assistant, our ML engineers:

🔸Loaded the parsed textual data into the vector database to be able to interact with it further and used an RAG approach to increase the accuracy of the LLM at the expense of facts extracted from external sources

🔸Performed feature engineering for the tabular data to improve the model performance and introduced an agent table processor as another LLM that helped us analyze the tables

🔸Implemented an embedding router for the system to determine the nature of the request and act accordingly

🔸Added an extra “general” type of request to cover general inquiries or tasks not specifically related to the knowledge base

BUSINESS OUTCOME

The information retrieval system proved itself a powerful tool to find and interpret information. The chatbot is fast and has reduced the time needed for search and information analysis from hours to minutes. As a result, the solution helps to save hours on operational tasks and maximize efficiency at the workplace.

TECHNOLOGIES USED

Python, FastAPI, Pandas, Scikit-learn, PyTorch, Transformers (HuggingFace), Streamlit, OpenAI, AWS Lambda, AWS S3, AWS DynamoDB, Open Source LLMs, Docker

ABOUT THE COMPANY

Our customer is a representative of Strategic Integrations Group from HubSpot — an owner of a cloud-based CRM platform. They aim to provide a best-in-class experience to teams by offering a developed net of integrated software solutions that meet all their possible needs and allow for inbound marketing, sales, and customer service.

BUSINESS CHALLENGE

The main purpose of the Strategic Integrations Group, with which the Intelliarts team collaborates in this project, is to provide employees of businesses with access to all their essential tools within the HubSpot ecosystem through custom integration development. This way, the team ensures that users stick with the HubSpot platform without the hassle of switching toolsets, making HubSpot a powerful, competitive, and truly comprehensive platform. So, they reached out to Intelliarts to get assistance with full-cycle web software development.

SOLUTION

Our engineers conducted the end-to-end development of services that would enable the integration of multiple 3d party software tools with the HubSpot CRM platform. The list of software tools to be integrated include: NetSuite, Microsoft Dynamics 365, Slack, Zoom, Mailchimp, OneSignal, Salesforce, Jira, SurveyMonkey, and Marketing Events Public API. 

We ensured that the resulting solution has the requested functionality and features, specifically:

🔸The developed services support the synchronization of data received through the many software solutions integrated with the HubSpot platform. It concerns, for example, contact details of Zoom meeting participants, session recordings, and other crucial information.  

🔸Services are being modified continuously to match the latest updates of corresponding software solutions. It ensures the high performance of integrations. 

🔸Provides access to integrated software solutions from HubSpot's digital environment and vice versa. Both automated data synchronization and manual information pulling are supported. Access and use are intuitive.

BUSINESS OUTCOME

The requested software integration has been released successfully. The Intelliarts team achieved the following results in this project:

🔸 Salesforce integration has more than 11 thousand active installs from the HubSpot marketplace as of now. 

🔸 The integrations developed for the HubSpot internal service team have more than 150 thousand active users. 

🔸 Spending of time and effort on business communication and data transfer have been optimized notably. It allows teams to get the most out of their tech stack and achieve better results from their operation. 

The Intelliarts team keeps providing continuous support and maintenance, as well as releasing updates to all of the developed services.

TECHNOLOGIES USED

Java 17, Dropwizard + Guice, AWS, Zookeper, Elasticsearch, Kafka, React, MySQL, HBase, Hadoop

ABOUT THE COMPANY

Our customer is ProvenBase — a US-based company that provides B2B diversity sourcing solutions. It’s committed to helping businesses find the most suitable talents, streamline the hiring process, and save resources on building teams through innovative technologies and diversity, equity, and inclusion (DEI) practices. 

BUSINESS CHALLENGE

The customer aims to have a fully-fledged B2B platform with an integrated AI module for job description parsing and matching candidate profiles with job requirements. This would allow customers of ProvenBase to streamline candidate sourcing. So, the customer reached Intelliarts to get assistance with building an ML solution that would solve this business task and integrating it into the existing SaaS product.

SOLUTION

Our engineers collected and prepared the dataset to be used for machine learning models training and chose metrics for the evaluation of models’ performance. Then our engineers built and trained NLP models for job parsing and candidate profile matching. We also built a semantic search engine that can search for profiles based on text input and support multiple search filters, like age, gender, racial origin, etc. Once an ML solution that meets customer requirements in terms of performance was created, Intelliarts engineers created an API to integrate it into the existing SaaS product. 

We ensured that the updated SaaS product has the requested functionality and features, specifically:

🔸 The platform can retrieve profiles of job seekers and provide key information regarding their candidacies.

🔸 A semantic search engine can match the input query with profiles and find candidates with the desired qualifications or specialization.

🔸 The search engine can filter candidate profiles based on chosen criteria, including professional background, specific skills, region of living, gender, or ethnicity.

🔸The AI module can automatically match job requirements with available profiles and range candidates based on their suitability for a position. 

BUSINESS OUTCOME

The AI solution shows over 90% accuracy for gender and ethnicity detection. The developed solution provides users with a set of diversity sourcing tools and complementary functionality that allows for quick, semi-automated matching of the job position requirements with profiles of candidates sourced from job boards or social platforms, job profile sorting, and meeting a company’s DEI goals easily. Now, ProvenBase’s product users, which supposedly are recruiters and HR specialists, can:

🔸 Get access to a diverse pool of talents.

🔸 Utilize analytical instruments and DEI resources.

🔸 Diversify their talent pipeline in accordance with the best DEI practices.

🔸 Minimize spending of time and financial resources on candidate sourcing and hiring.

Successful integration of the AI-powered module into the product allowed the company to be better prepared for the industry challenges and the high competition and move to the next investment round. For the next round of product enhancements, we advised the customer to further improve the semantic search by adding the possibility to select candidates based on a particular experience, e.g., narrowly focused projects, specific IT architecture, etc. 

TECHNOLOGIES USED

Python programming language, SentenceTransformers framework, Beautiful Soup Python library, Docker, Hugging face Hub, Label studio data labeling tool.

ABOUT THE COMPANY

Our partner specializes in selling modern Electric Vehicle (EV) charging solutions with premium 24/7 customer service, among which there are turnkey certified charging stations, a trusted EV driver mobile application, a platform for EV chargers management, and a variety of other services. ​​Since Intelliarts has already been a trusted partner for software solution development for this company for multiple years and since the company always strives to improve their user experience, the customer reached out to us with another business challenge to solve.

BUSINESS CHALLENGE

Just like any other devices, EV charges are at risk of abrupt outages, which causes lots of inconvenience for the company’s end-users. Aiming at improving the customer experience, as well as reducing downtime, the company wanted to implement a predictive maintenance (PdM) solution to be able to predict and prevent failures of their charging stations.

SOLUTION

Intelliarts has conducted a detailed analysis of historical data received from EV chargers via OCPP protocol and data anomaly analysis as the first steps toward building the ML-powered solution for predicting anomalies in the behavior of EV charging stations. Precisely, our team of data engineers:

🔸 Discovered the ambiguity in the existing EV charging station health state labeling. So, we advised the customer on how to improve and automate the labeling process.

🔸 Received multiple insights into the quality of the state diagnosis data and the extra data required but missing to build the PdM solution we planned for. We provided the customer with specific recommendations on how to increase the data quality and predictive power by preserving the raw historical data instead of aggregated one and by using the proper format and type of data.

🔸 Implemented an anomaly detection solution by using three unsupervised algorithms, namely DBSCAN, the ​​isolation forest algorithm, and the local outlier factor, to better understand which anomaly behavior usually results in EV charger outages.

🔸 Together with the station's technical experts, we performed a series of discussions to interpret the insights collected during data and anomaly analysis from the business perspective, so the company could improve the behavior of their EV charging stations based on this.

BUSINESS OUTCOME

As a result of the current stage of the PdM project implementation, our partner has obtained a detailed recommendation on how to improve their data collection pipeline in order to build the desired predictive model(s) at the next project stage. Specifically, we have advised on the extra raw data that needs to be collected; the proper format and type of data to use; the automation of data labeling; and how to build proper cold storage on top of AWS S3 for effective long-term storing of historical data.

Now our software development engineers are collaborating with the customer’s team to implement those recommendations and collect more historical data in the suggested format so we could move forward with building the ML-powered PdM solution as soon as the quality of data improves.

TECHNOLOGIES USED

AWS SageMaker, AWS Glue, AWS Athena, AWS QuickSight, AWS S3 Glacier, Metabase, MongoDB, Kafka, Scikit-learn, XGBoost, Pandas, Matplotlib, Plotly, Docker

ABOUT THE COMPANY

Our customer, a large manufacturing company, has a complex hydraulic system as a part of their facility, and they care a lot about its proper predictive maintenance strategy. The company sees this as a way to significantly reduce their maintenance costs as well as to improve productivity and lower the probability of unexpected downtime.

BUSINESS CHALLENGE

The company reached out to us in search of a solution for the repeated malfunctioning of the specific components of their hydraulic system (cooler, valve, pump, and hydraulic accumulator). The four components have broken down periodically, disrupting the whole hydraulic system and stopping the operation of the entire facility.

SOLUTION

Within the scope of our cooperation, Intelliarts has implemented a machine learning-based system that could predict the degradation level of specific components with a high level of accuracy. The solution was composed of four separate machine learning models for each of the components with an average prediction accuracy of 98%. During the project, we

🔸 Processed the records from IIoT sensors in the dataset and conducted a detailed data analysis to make sense of the data by uncovering trends and patterns.

🔸 Solved the problem of model overfitting, which was caused by a big amount of features in the system. It was important to prevent this problem, so it couldn’t affect the model’s ability to generalize.

🔸 Chose the XGBoost classifier, StratifiedKFold, and RandomForestClassifier algorithms for the ML models as these have shown the best results on the available data.

🔸 All these actions allowed us to increase the prediction accuracy on test sets to 98% for each of the four ML models.

🔸 We also developed API endpoints, triggered after each load cycle of the system. This should help the customer with the model monitoring to identify when the performance drops, and the model requires retraining or tuning.

🔸 Together with the management of the company, we also conducted training sessions to educate the personnel on how to use the ML solution in their day-to-day procedures.

BUSINESS OUTCOME

At the end of the project, we provided the customer with the efficient ML solution that could help the manufacturer with predicting malfunctions of the components of the hydraulic system and, thus, ensure proper maintenance of their equipment. The solution proved to be almost 100% accurate in its results. The Intelliarts team continues to keep in touch with the company so, if they notice any significant changes in the data, we’re always ready to assist and retrain the models.

TECHNOLOGIES USED

AWS SageMaker, AWS Lambda, Matabase, Scikit-learn, Dask, Pandas, Matplotlib, Seaborn, TPOT, Docker

ABOUT THE COMPANY

Our customer is a midsize insurance business specializing in property and car insurance. This is also the company that Intelliarts has collaborated with for a long time. We know our partner as the company truly committed to improving their operational efficiency and cutting costs. So, the customer is used to constantly monitoring new ways to optimize their operations, especially with the help of innovative technologies.

BUSINESS CHALLENGE

Cold calling is one of the key strategies for how the company attracts new clients. However, phone numbers used by the company’s insurance agents are periodically flagged as spam, and the insurer suspected this could affect the effectiveness of their cold calling sales channel. The company reached out to Intelliarts to test this hypothesis by using a data science approach and help them solve the problem of the low cold calling success rate.

SOLUTION

We completed a series of data analyses to summarize the main characteristics of the phone call data, discover patterns, and test assumptions. Our main conclusion was that the underperformance of the customer’s cold calls proved not to correlate with their phone numbers being marked as spam. Specifically,

🔸 This was confirmed by the general phone call analysis, which helped us check how the amount of successful cold calls grew over time and whether there was any correlation between effective cold calls and all calls generally at the moment when the numbers were flagged as spam.

🔸 We reached the same conclusion with the number migration analysis. During this research, we monitored the effectiveness of cold calling over time and expected the performance to drop after the number was marked as spam. This did happen, but occasionally the performance grew up again after a few more calls were made from the same number, proving no tight correlation between the two events.

🔸 During the underperformance vs. spam analysis, the Intelliarts team contrasted the low performance of a particular number to spam detection. We discovered that the low success rate of a specific phone number didn't correlate with the same phone number being marked as spam later. As a result, we ensured that while marking phone numbers as spam did provide extra management efforts related to phone number management, it didn't have any direct influence on the overall effectiveness of cold calls.

Despite having no evidence suggesting that the underperformance of cold calls was caused by the company’s phone numbers being marked as spam, we discovered other factors that mattered. We found that the agent who was calling, the demographics of the lead the company was calling to, the timing of the original and follow-up calls, and the strategy of how the cold calls were managed affected the cold call success rate.

BUSINESS OUTCOME

The comprehensive data analysis that Intelliarts conducted helped us form a set of recommendations for the customer with further action needed to solve their business challenge. We advised the company to continue investigating other reasons that impacted cold call performance since the hypothesis about phone numbers being marked as spam wasn’t confirmed.

We built the MVP for this project, which, in the long run, can help the customer detect which factors specifically cause the low performance of cold calls and avoid such patterns. Now the customers’ insurance agents are testing the MVP solution in the fields, so once we receive the initial feedback, we will proceed with the second phase of building the full-scale ML solution for phone number management.

TECHNOLOGIES USED

AWS SageMaker, Metabase, Scikit-learn, NumPy, XGBoost, Pandas, Matplotlib, Snowflake, Docker

ABOUT THE COMPANY

Our customer (under NDA) is a global manufacturer of home appliances. The manufacturer is well-known for their commitment to quality and safety standards, and that’s why the company monitors every step in their production process. They thoroughly record all the sensor measurements for each module along the production line to use this data further to optimize manufacturing operations.

BUSINESS CHALLENGE

To improve their production line performance, the customer needed to solve the problem of repeated and unexpected equipment failures in their production line. Because of this issue, the company suffered from costly expenses on new equipment components and materials for repair or rewind; labor costs; freight rate to transport the new parts; and downtime. Besides, inconsistent work in the production line caused delays in shipping and affected overall plant productivity.

SOLUTION

We’ve built the machine learning (ML) solution to predict system failures ahead of time and reduce breakdowns to a minimum acceptable level by performing necessary cost-effective predictive maintenance.

While building the solution, the Intelliarts team faced a couple of challenges, such as the need to process and analyze huge historical and real-time datasets collected from the IoT/IIoT sensors. We dealt with this problem by building the scalable data processing unit on top of Dask and deploying it to the AWS Fargate cluster. This solution allowed us to operate big volumes of data in a scalable and effective manner.

To increase the predictive power of ML algorithms used in the solution and improve the model performance, our team of engineers:

🔸 Encoded categorical data using one-hot encoding

🔸 Clustered samples by production flow using k-Means

🔸 Reduced the number of numerical features that were highly linearly dependent

🔸 Calculated lag features as one of the most important feature types for the solution 

The next step was to achieve the best results in modeling. We experimented with different algorithms and proofs of concepts and finally chose the Extreme Gradient Boosting classifier after getting the best score in this case and increasing its prediction performance with model tuning. After deploying the prediction model, our team defined policies and helped the management of the company to teach the personnel how to use the solution in their day-to-day work.

BUSINESS OUTCOME

The model is already deployed to the manufacturer’s system. Since it was a full-cycle data science project, we also set up a monitoring system and built dashboards so the company could track the model results and reach out to us in case of any significant changes in data or the model performance. The manufacturer is completely satisfied with the results — the model showed over 90% accuracy in equipment failure prediction, which has already helped the company reduce their maintenance costs by 5%.

TECHNOLOGIES USED

AWS ECS, AWS Fargate cluster, AWS SageMaker, Metabase, Dask, Pandas, NumPy, Scikit-learn, XGBoost, PyTorch, Docker, Neptune.ai, Seaborn