Roman Kriuchko

Built Foxquant, an AI-driven quantitative trading platform. Led the design of LLM-assisted research workflows, real-time signal models and the backtesting engine powering strategy discovery for retail and pro traders.

• Designed and shipped LLM-based copilots that translate natural-language hypotheses into runnable trading strategies

• Built a multi-agent research stack (LangGraph + OpenAI / Anthropic) for market analysis, news ingestion and signal generation

• Architected a hybrid RAG layer over financial filings and time-series data using pgvector and embeddings re-ranking

• Trained and fine-tuned forecasting models in PyTorch and integrated them with a low-latency FastAPI inference service

• Stood up MLOps pipelines (Docker, GitHub Actions, model registry, eval harness) for safe, repeatable releases

• Owned the end-to-end Next.js + Python product surface used by early-access traders

Built AI tooling for blockchain security and analytics. Combined LLMs, retrieval and traditional ML to triage smart-contract risk, automate research and surface on-chain insights.

• Built an LLM-powered smart-contract review assistant that highlights vulnerability patterns and explains them in plain English

• Designed RAG pipelines over audit reports, EIPs and protocol docs with hybrid search (BM25 + embeddings)

• Developed ML models (Python, scikit-learn, PyTorch) for anomaly detection on on-chain transaction graphs

• Built internal Django + FastAPI services exposing model inference and agent workflows to the security team

• Partnered with research to evaluate prompts, tune retrieval and ship guardrails (PII, prompt-injection, output validation)

Combined applied ML with blockchain product work — from NLP-driven chat experiences for crypto wallets to smart-contract automation and on-chain data analytics.

• Built an NLP-powered wallet chat that lets users discuss transactions and gets context-aware suggestions from an LLM backend

• Trained classification and ranking models for NFT collections, used in recommendation and discovery surfaces

• Designed and deployed smart contracts and the Python services that orchestrated minting, indexing and analytics

Led AI feature work inside EdTech and SaaS products: chatbots, semantic search, content generation and personalization for products used by tens of thousands of learners.

• Shipped an AI tutoring chatbot (LLM + retrieval over course content) that increased lesson completion by 30%

• Built recommendation and personalization services in Python, integrated with the React / Next.js learning UI

• Mentored engineers on prompt engineering, evaluation harnesses and safe rollout of AI features

Worked on interactive 3D and AR products with computer-vision components: object tracking, segmentation pipelines and ML-assisted asset generation.

• Implemented computer-vision pipelines (OpenCV + classical ML) for AR object tracking inside training simulations

• Built a chatbot for customer support that reduced response times by 50% using intent classification and templated answers

• Optimized model inference and asset pipelines, cutting load times by 40%

Joined an applied research team on non-invasive cancer-detection algorithms. Picked up deep learning fundamentals and built tooling around medical image processing.

• Implemented and benchmarked image-processing and ML algorithms for medical image classification

• Built internal data-labeling and evaluation tools using Python and web technologies

• Co-authored research write-ups and contributed to publications in image-processing venues