Services
Full-cycle product development
We build web applications and platforms end to end: discovery, architecture, development, testing, release, and support. The focus is AI in the product—search, support, analytics, automation, and new AI-powered features.
AI integration for product and operations
We pick use cases with measurable impact (support, search, analytics, automation) and implement them. We set up data and integrations, define metrics, and put quality control in place so results show up in business KPIs.
R&D: prototyping and hypothesis validation
We validate the hypothesis in practice: define success criteria, build a prototype, measure quality, and map risks. You get a metrics report and an implementation plan. If the criteria are met, we move to MVP and integration.
Expertise
Natural Language Processing
We build AI components for products and operations—from problem framing to integration and quality metrics.
- Multi-agent systems: orchestration of tools and roles
- MCP for agentic workflows
- Quality control and predictable behavior
- NLP tasks: classification, sentiment analysis, summarization
RAG & knowledge graphs
We enable search over internal data and source-backed answers so knowledge stays reliable and up to date.
- RAG pipeline: indexing, retrieval, reranking
- Source-backed answers and hallucination control
- Knowledge graphs: entities, relations, queries, updates
LLM inference optimization
We reduce latency and cost by profiling bottlenecks and speeding up production pipelines.
- Profiling and bottleneck analysis
- Caching, batching, and request optimization
- Model selection for quality, speed, and cost
CPU/GPU/edge deployment
We adapt models and services to different hardware and deployments: packaging, configs, and monitoring.
- Tuning for CPU/GPU/edge scenarios
- Packaging and deployment
- Operations: monitoring and updates
Data for models
We build a managed data loop: collection, cleaning, labeling, and dataset versioning for training and fine-tuning.
- Data collection and cleaning
- Labeling and QA
- Prep pipelines and dataset versioning
Classical ML, CV, Recommenders
Production ML beyond LLMs: forecasting, CV, recommender systems, and audio—with clear metrics and reliable operations.
- Classical ML: boosting, anomalies, clustering
- Computer vision: detection, segmentation, OCR
- Recommenders: ranking, personalization, A/B tests
- Audio: ASR, diarization, event classification
Who we are
We’re a team of graduates and educators from HSE University, Saint Petersburg State University, and ITMO University. Researchers and practitioners with experience across leading Russian and international IT companies. We take business goals end-to-end: clarify the objective, build the solution, ship it, and measure the impact.
Projects
MARS
2025
MARS is a platform for multi-agent system experiments on CTF benchmarks, enabling agent tracing and evaluation. It supports composition of multiple agents with handoff and collaborative task solving. Includes two modes: full testing on virtual machines and fast inference tests for development. Integrates RAG for information retrieval, tools for environment interaction (SSH, command execution), and a tracing system for analyzing agent behavior. Configuration via declarative files allows customization of experiments, agents, their tools, and evaluation metrics without code changes.