Companion-AI Platform

Role: Senior AI Solutions Engineer

Architected and developed from scratch an enterprise-grade multi-tenant AI platform serving specialized business agents (Legal, PM, HR, Finance). Built a comprehensive RAG pipeline with hybrid embeddings and opt-in LLM/ColBERT re-ranking, dual vector databases, and semantic search to balance accuracy and speed. Designed a flexible integration framework with Finlex (Finnish legal database) and SharePoint to provide composed retrieval across multiple sources, supporting augmentation from external data and internal customer systems. Implemented advanced RBAC with Zanzibar-style permissions, a micro-frontend architecture using Webpack Module Federation, and containerized deployment enabling customer-specific service selection. Implemented LLM and Tool registries to power multi-purpose, multi-agent workflows with strict multi-tenant isolation.

April 2025 – September 2025

Public Procurements AI

Role: Senior AI Solution Developer

Designed complete system architecture from concept to production. Developed Python backend (Azure Functions, AsyncIO, Slack Bolt) and Next.js/TypeScript frontend with responsive design. Implemented ML pipeline with BGE-M3 embeddings, RRF hybrid search, and GPT-4o integration. Built zero-trust security with Azure Key Vault, comprehensive testing strategy (unit, integration, E2E with testcontainers), and automated CI/CD pipelines. Created user feedback system, real-time notifications, and monitoring. Deployed infrastructure as code (Bicep) with auto-scaling serverless design. Delivered end-to-end solution enabling intelligent RFP automation and business opportunity discovery.

May 2025 – August 2025

Serverless Embedding Service

Role: Senior AI Solutions Engineer

Architected and developed complete serverless ML infrastructure from scratch. Built FastAPI service with async processing, batch optimization, and comprehensive error handling. Implemented cold start optimization achieving 70-80% performance improvement through Azure Storage mounts and Premium ACR. Created automated deployment with Azure Developer CLI, Bicep templates, and container registry integration scripts. Developed model loading strategies, GPU optimization, and multi-embedding type support (dense/sparse/ColBERT). Built monitoring, health checks, and production-ready logging. Delivered scalable ML service with enterprise security and zero-downtime deployment capability.

April 2025 – May 2025

Personnel Development AI Assistant (Slack App backend)

Role: Senior AI Solutions Engineer

As sole architect and developer, I designed and implemented this competence development assistant using AI-assisted coding techniques.

I created the event-driven architecture with an asyncio-based event bus implementing publish/subscribe and request/response patterns. I architected the agent system using Microsoft's Autogen framework, developing SwarmGroupChat functionality with proper handoff between specialized competence development agents. I built a comprehensive component system with registry, factory, and caching mechanisms for extensible management. I implemented the repository pattern for standardized database access across all data models, ensuring consistent handling of development plans and user profiles. I created testing infrastructure including mocks, integration tests, and performance benchmarks.

March 2025 – April 2025

R&D: Autonomous Multi-Agent AI Trading System (PoC)

Designed and implemented a multi-agent trading system using LLM-based agents for analytics, trade execution, risk supervision, and sentiment filtering.

Deployed local and cloud LLMs with automatic fallback and latency-aware routing; integrated real-time sentiment analysis using a custom-trained topic classifier on Hugging Face (FinNews Topic Classify).

Built a semantic enrichment pipeline with dense (BGE-M3) and sparse (SPLADE/BM25) embeddings, served via Infinity embedding engine for high-throughput, low-latency vectorization.

Engineered a Polars-powered indicator engine for real-time RSI, SMA, and earnings surprise tracking, generating market context embeddings for retrieval and filtering.

Integrated Qdrant vector database to support hybrid queries over news, market context, and entity metadata; supported filters by symbol, sector, sentiment, and events (e.g., EPS, M&A).

Implemented Prometheus + Grafana observability, tracking token usage, P&L performance, execution latency, and agent loop metrics.

Established automated safety and feedback loops:

• Risk halting on daily drawdown or LLM failure
• Backtests with dynamic strategy weighting using reward feedback and multi-agent consensus

Solo Research & Engineering Project | March 2025 – Present

Financial News Topic Classifier (Published on Hugging Face)

Developed a financial topic classifier using transfer learning by adapting the Trading-Hero-LLM (originally trained for sentiment classification) to a new task: multi-topic categorization of financial news.

Froze pretrained layers and fine-tuned classification heads on Kaggle's ZeroShot Financial News dataset covering 20 business topics (e.g., Earnings, Fed, M&A, Commodities).

Engineered the label space and input representation to align with single-label classification and ensure class separation in imbalanced conditions.

Achieved macro F1 > 85% with consistent per-class performance across highly varied financial narratives.

Published the model to Hugging Face, making it usable for news categorization, finance-aware NLP pipelines, and downstream routing/logging in trading systems.

Integrated the classifier into an LLM-agent pipeline for topic-aware trade signal filtering and RAG document tagging.

This project demonstrates practical transfer learning, financial domain adaptation, and open deployment of reusable NLP components.

Finance NLP Model Development & Transfer Learning | 2025 | huggingface.co/leonas5555/finnews-topic-single-classify

Cloud Product Solutions: Online Compliance & Accreditation Management (TL)

Played a key role in the end-to-end development, continuous maintenance, enhancement, and technical support of the solution. Acted as the Technical Team Lead for the solution-focused team comprising full-stack developers, QA engineers, and operations staff. Responsibilities included collaborating with customers to plan sprints and project deliverables, scheduling and coordinating team activities, strategizing technical implementation, designing event-driven orchestration and high-availability features, hands-on programming, DevOps pipelines, and Azure Kubernetes components. Led the team to deliver a scalable, high-quality solution while driving technical excellence and ensuring alignment with business objectives.

Confidential - Global - Legal/Public sector | December 2023 - January 2025 | Technical Team Lead

Cloud Product Solutions: Online Compliance & Accreditation Management (SWD)

Full-Stack Development: Engaged in hands-on programming across both frontend and backend components. Developed and maintained robust microservices, including Java Spring Boot and Kotlin Ktor API services. Implemented Apache Kafka consumers and producers to streamline data processing. Designed and executed efficient data storage and retrieval solutions using PostgreSQL, Elasticsearch and Redis in a microservices environment. Crafted dynamic SSR React.js UI components. Established and configured DevOps pipelines within Azure; managed Azure Kubernetes and platform components using Helm and Terraform.

Confidential - Global - Legal/Public sector | January 2022 - December 2023 | Senior Software Developer

Cloud Product Solutions: Online Learning Platform Migration

Engineered a seamless transition from Oracle DB to PostgreSQL in Azure SQL by designing and implementing robust data extraction and migration processes. Developed and optimized migration scripts using Docker and Azure-based cloud solutions. Built and maintained DevOps pipelines and Kubernetes Helm charts. Developed a microservice integrating PostgreSQL and Elasticsearch for robust data management and high-performance search. Developed Java Spring Boot microservices and SCIM API to synchronize user HR and learning data from third-party vendors, and created Liferay UI portlets to enhance user experience.

Confidential - Global - Travel Industry | June 2021 - December 2021 | Software Developer

Cloud Product Solutions: Global Search

Developed a feature-rich search UI using React.js and Redux, integrating advanced pagination, facets, sorting, dynamic results prompts, and intelligent text suggestions. Engineered scalable Kubernetes backend API microservices that process UI actions and deliver paginated search results directly from the Elasticsearch API. Configured Kubernetes Helm charts to streamline the deployment of related services and SSR UI components.

Internal Product | June 2019 - October 2019 | Software Developer