Dmytro – AI, Python, PyTorch

Resemolnet is an AI-powered dynamic pricing and demand forecasting system for optimizing weekly tour prices. It processes booking history, product metadata, and operational constraints to forecast demand, estimate price elasticity, and recommend revenue-maximizing prices. The pipeline automates validation, preprocessing, forecasting, optimization, and output generation, producing price schedules for hundreds of tours while managing urgency adjustments and overbooking scenarios.

• Lead the AI part of the project, including new feature development and maintenance/bug fixing
• Perform model research and selection for upcoming functionality
• Drive the new Bareboats module, work with new data types, and define the ML approach
• Maintain and update the existing system, including implementing daily update pipelines
• Run testing and research based on client requests, covering both ML tasks and domain questions

Accomplishments:

• Delivered an automated weekly pricing system
• Increased revenue potential by aligning prices with forecasted demand and sales targets
• Reduced manual pricing work through automation
• Designed a scalable architecture to support future real-time optimization
• Improved system resilience via overbooking detection and urgent inventory handling

Create a virtual agent to handle FAQs and log tickets in the TechSupport system. Extract names, emails, phone numbers, issues, and details during the conversation, fill the ticket in the system, or send a self-guide.

• Design conversational flow
• Develop a Rule-Based chatbot and tweak parameters for upgrading conversational quality
• Create a system for managing connections to the TechSupport system for creating tickets
• Extract detailed information from a conversation with the user
• Post-process conversation records and text for extraction
• Implement an issues classification system for smarter assignment of tickets on the TechSupport side
• Integrate OpenAI Whisper for improving sound processing

Accomplishments:

• Improved customer support efficiency by building a chatbot that captures structured issue details and automatically creates and routes tickets, reducing manual triage and speeding up resolution.
• Increased routing accuracy and conversation quality by combining rule-based logic with issue classification, post-processing, and enhanced audio understanding via Whisper integration.

Web platform that automatically captures news and press releases related to copper stocks in the commodities market, transforms this information using AI into professional journalistic articles, and publishes them automatically.

• Create a processing script for Amazon Textract for collecting and filtering input text
• Implement and tune the OpenAI model
• Upgrade tokenization for the model to improve the speed of processing
• Prompt engineering for generating and summarizing text from the articles
• Create a post-processing module
• Logging and solving bugs in the service

Accomplishments:

• Accelerated document processing by integrating Amazon Textract with an optimized OpenAI pipeline, improving throughput and reducing end-to-end latency via upgraded tokenization.
• Increased generation and summarization quality through targeted prompt engineering and robust post-processing, while boosting production reliability with strong logging and rapid issue resolution.

An AI-powered conversational assistant designed to conduct pre-election political surveys in Australia. The bot introduced users to all candidates and explained their positions, then asked users how they felt about each candidate’s views.

• Design and refine the conversational flow for smooth interactions
• Build and configure the chatbot end-to-end
• Optimize bot settings for realistic voice and accurate extractions
• Implement fallback logic for unexpected user inputs
• Run demos and perform regular testing to ensure quality

Accomplishments:

• Increased user engagement and reduced drop-off rates by transforming the chatbot into a more natural and goal-oriented conversational experience
• Improved overall reliability and response accuracy through continuous testing cycles and iterative enhancements, resulting in a stable solution ready for deployment at scale.

Built a system for clustering and summarizing news texts using embeddings, deep learning, and topic modeling techniques. The pipeline combined OpenAI/Nomic embeddings, autoencoder-based dimensionality reduction, and HDBSCAN clustering.

• Working on text embedding, classification, and summarization tasks
• Training, testing, and tuning classical ML models and some neural nets on PyTorch
• Conducted research-focused tasks, evaluating various approaches and models, and optimizing solutions to work efficiently with limited hardware resources
• Developed and deployed small RESTful API services for seamless integration into the overall product
• Collaborated in a small team to maintain and enhance legacy text analysis services on a daily basis

Accomplishments:

• Enhanced text analytics capabilities by improving embedding, classification, and summarization quality, achieving higher model accuracy while keeping inference efficient on limited hardware resources
• Accelerated experimentation and deployment cycles by optimizing ML pipelines and delivering lightweight RESTful APIs, while ensuring stability and continuous improvement of legacy NLP services in a fast-paced team environment.