AI and ML-based solutions
— Machine Learning
— Deep Learning
— Convolutional Neural Networks
— Sentiment Analysis
— Natural Language Processing
— Facial Recognition
— ChatGPT integration solutions
— Chat-bots
— Model fine-tuning (Google's T5)
— Text-to-Image models (DALLE, Stable Diffusion)
An application with the voices of over 40 celebrities, mimicked by artificial intelligence. Users record audio messages and share them on social networks.
The neural network was initially trained on a huge amount of data (more than 30 hours of speech recordings) of one particular speaker who is not a famous person. Further training on the data of the selected celebrity lasts no more than 2 hours. In essence, the neural network has "its" voice and overlays the intonations and speech features of the person being parodied on top of it.
Технологии синтеза упакованы в мобильное приложение на основе React Native.
In 3 months we built an MVP and managed to occupy a new niche. Saved 50% development budget by choosing cross-platform technology and speed. At the moment the application has been downloaded by more than 9 million users.
Product for working with records of past video conferences within corporations Solves the problem of chaos in corporate video calls. A simple catalog with video recordings is inconvenient for searching important recordings when you need to bring up the context of a meeting or recall the terms of arrangements.
A product with extensive functionality has been created:
- Ability to recognize participants
- Audio transcription
- Keeping general notes for video
- Search for video recordings and their key points: participants of the recording, speech start time, trigger words, emotional peaks in the conversation
The task was to improve the security of corporate devices even in remote environments. We offered a new approach to protect sensitive information by ensuring that only an authorized user can access their device using advanced facial recognition.
We designed a facial anti-spoofing system in C++ using OpenCV to prevent fraudulent biometric access. The solution employed a fine-tuned DNN from OpenCV's model zoo to differentiate genuine from spoofed facial samples. Implemented C++ optimization techniques, such as Quantization and pruning, to speed up DNN inference. Multithreading was utilized for simultaneous processing of multiple video streams. GPU acceleration further reduced latency, ensuring swift and reliable spoof detection.
Cash currency detection system to prevent fraudulent currency substitution operations.
The developed system can detect currency by video stream and reconcile these events with the bank's automated transaction accounting system, controlling fraudulent activities. In case of discrepancy, such transactions are highlighted to the security service for further proceedings.
Implementation of the system helped to save up to 90% of time for the bank security service when searching for fraudulent operations.
The ATM can detect gender and age, detect objects, emotions, gestures, recognize faces, and is equipped with a contactless control system. Testing of the solution in a partner café showed a 5% increase in the conversion of unique visitors into sales.
We worked on continuous tracking based on skeleton detection and pairing it with a person's face, so that large number of people near the booth didn’t interfere in image processing.
Our team designed and developed an intuitive non-contact interface for interaction with the booth, prevented situations when finger gestures are poorly read due to distance. Also we optimized an ensemble of neural networks to achieve real-time display of high resolution pictures on the screen.
Customer emotion scoring system to determine service satisfaction level. A turnkey solution automatically analyzes service quality and customer emotions from the video stream using ML and CV.
As a result of implementing the solution in the pizzeria chain, an increase in customer satisfaction of more than 30% was recorded and a fairly strong correlation (0.86) between customer satisfaction and the size of the average check was found.
Video-based feed quantity detection system for dairy farms. Utilizes the customer's existing video surveillance infrastructure to determine the presence and volume of feed.
The neural network classifier is trained not only to understand the presence of feed in a given area, but also to accurately determine its quantity by 4 classes, for prompt delivery.
We needed to create a chatbot that could accurately answer customer questions using existing documentation and frequently asked questions.
To solve this problem, a variant of BERT, finalized on the basis of the information bank, was used as an engine for the chatbot.
As a result, the chatbot was able to handle over 90% of user requests, reducing response times and increasing customer satisfaction.
Google T5 was used to create a bot for answering mathematical questions, refined from the "chain of thoughts" dataset, using "zero-shot" and "multiple-shot" techniques.
Google T5 was chosen as the engine for this chatbot, which was further refined based on the "chain of thoughts" dataset. With the added benefit of developing zero-shot and few-shot prompts, the bot was able to solve and explain over 95% of math problems, which proved invaluable to its target audience.
A system to speed up routine processes when hiring IT specialists. Automates candidate selection, skills testing and communication.
AI analyzes resumes + gives feedback to candidates. If necessary, it requests additional information. Next, candidates pass to the testing stage. The AI module checks the candidates' code and assigns a grade without techlid. Candidates go through the auto-testing phase. The AI evaluates the answers and validates or rejects the candidate. The bot coordinates interview dates. Techlead receives validated candidates for interviews.
Smart system for optimizing the HR management process A tool for effective management of companies with more than 100 employees. We develop a tool that can collect and analyze data on company structure and employee performance – then produce a predictive model of changes in the company that will help improve its profitability and efficiency.
The system provides effective management of the company:
- collects and analyzes data
- predicts metrics
- recommends control actions through ML modules
It is possible to create a full-fledged digital model of the company, which will reflect the relationships of structural units with managerial and financial metrics.
of project implementation
Vacancies
Musthaves:
- Proficiency in one or more programming languages commonly used in AI/ML, such as Python, R, Java, C++, or Julia.
- In-depth understanding of various machine learning algorithms, including supervised learning, unsupervised learning, and reinforcement learning.
- Familiarity with deep learning techniques and frameworks like TensorFlow, PyTorch, or Keras.
- Strong skills in data manipulation libraries like NumPy and pandas for data preprocessing and analysis.
- Ability to create meaningful visualizations using libraries such as Matplotlib, Seaborn, or Plotly to communicate insights from data.
- Proficiency in feature engineering techniques to extract relevant information from raw data.
Additional skills:
- Proficiency in version control systems like Git for collaborative development.
- Knowledge of SQL and NoSQL databases for managing and querying data.
- Strong foundation in linear algebra, calculus, probability, and statistics, as these underpin many ML algorithms.
Responsibilities:
- AI/ML Model Development: Lead the design, development, and deployment of advanced machine learning models and algorithms to solve complex business problems. This includes selecting appropriate frameworks, libraries, and tools.
- Data Preprocessing: Implement robust data preprocessing pipelines, feature engineering techniques, and data augmentation strategies to ensure high-quality input for machine learning models.
- Model Training and Optimization: Train and fine-tune machine learning models using large datasets, and implement optimization techniques to achieve optimal performance in terms of accuracy and efficiency.
- Model Deployment: Take ownership of deploying machine learning models into production environments, ensuring scalability, reliability, and real-time performance.
- Algorithm Selection: Evaluate and select appropriate AI/ML algorithms and techniques for specific use cases, taking into account factors such as data size, complexity, and business objectives.
Musthaves:
- AI/ML tools : ALteryx, Qlik view, Python, Matlab, SQL, Tableau, R, Angular, GCP, AWS, Mach1 ML or any similar tools
Additional skills:
- Effective Communication
- Good coding skills
- Project Management
- Stake Holder management
- English level – B2
Responsibilities
- Work hands on and actively partner across the analytics chain - problem definition to resolution building including deployment in Ford Process
- Utilize industry standard AI/ML tools - optimization, database, visualization; recommend new tools / approaches as needed; build / guide building of tools where needed.
- Partner with cross teams as well as Global/regional teams at operating / leadership levels as needed across the Ford ecosystem as relevant
- Drive projects in a complex ambiguous business environment
- Manage channels of communication up and across as appropriate - POCs, simulations, presentations, etc
Musthaves:
- Ph.D. degree in electrical engineering, computer engineering, or computer science or Master s Degree (or equivalent) and at least 5 years research work experience
- Master s Degree (or equivalent) and at least 5 years research work experience
- Experience in developing algorithms for machine learning and deep learning, computer vision, and 3D scene.
- Intimately familiar with computer graphics, video processing systems, as well as corresponding software in consumer and professional products.
- Proficient in Matlab, C/C++, OpenCV, and Python (Tensorflow/Pytorch) programming applied to computer vision is required.
Additional skills:
- Experience in filing patents of newly developed technologies.
- Hands-on experience in applying Computer Vision and Machine Learning algorithms to real-world problems
- Proven track record of successful research accomplishments, published papers, and/or patent applications in the field of video encoding and processing.
Responsibilities:
- Works with a general understanding of competitive technologies in the area of focus.
- Works to understand the system context in which new technologies will be used and the requirements that the technology must fulfill for success.
- Creates early-stage conceptual models that demonstrate feasibility.
- Document and present the new architectures and algorithms in various forms, such as technical white papers and internal meeting
- Maintain the highest technical and moral integrity for Dolby in the workplace and marketplace.
Musthaves:
- Strong coding ability and familiarity with Python R programming languages
- Relevant experience, passionate about applications in data:crunching billions of datapoints for statistical modeling,data mining for insights and recommendation solutions
- Worked with ML data mining toolkits like NLP,Semantic Web,R,Core NLP,NLTK etcInformation retrieval libraries like Lucene SOLR fast paced,test driven, collaborative and iterative programming env
Additional skills:
- Good in communication
- Good in Team handling
Responsibilities:
- Build predictive models, develop advanced algorithms that extract and classify information from large datasets quantify model performance
- Evaluate emerging technologies that may contribute to our analytical platform Identify and exploit new patterns in data using various techniques
- Work with product managers and engineers to deploy models into production
- Good to have the knowledge in Insurance, Banking and Capital markets domains
