Viktor Andriichuk

Founder

Micro-SaaS studio run AI-first: research, code, payments setup, legal pages, support drafts and outreach go through one AI system, with decisions kept by a person. andriichuk.com

• AI in Order: productised service that sets up a personal AI operating system for owners who run several clients. Paid audit, then setup and optional care. aiinorder.com
• ConvertQBO: converts bank CSV, Excel, QFX, OFX and QIF files to QuickBooks .qbo in the browser, for bookkeepers and QuickBooks users. convertqbo.com
• Chrome extensions: Fast Chrome Screen Capture, Quick Chrome Notes and Ama Citation Generator, with optional paid plans.
• Also: tostranger (anonymous letters to strangers abroad, live); Faroio (ad-account observability for media buyers) and Sottocue (AI meeting assistant for macOS) in early access.

Founder

Romanian company running five businesses: DataFlux Software (AI voice agents and chatbots), Arctura (AI player support for iGaming operators), Intellectum Lab (AI automation for regulated finance), Mint Wave Media (content, SEO and performance marketing) and JLR Service (Land Rover specialist, Baia Mare). gazolin-production.com

Intellectum Lab — AI for EU regulated finance (2025 — Present)

• DORA operational resilience: continuous-monitoring layers on top of existing GRC tools; automated review of SaaS and LLM vendors against DORA Article 28; RAG-backed incident classification against DORA reporting thresholds.
• EU AI Act compliance: risk classification under Article 6 and Annex III; governance documentation for Articles 9–17; monitoring of model behaviour and drift with audit-ready logs; vendor exit and portability reviews for LLM deployments.
• Production RAG for regulated knowledge bases: policy, regulation and contract repositories with per-document source attribution, role-based access and full query audit trails.

DataFlux Software — voice AI for support and sales

• Milina (New York hospitality, voice ordering): LiveKit, Deepgram Nova-3, GPT-4o-mini and Cartesia; $0.09 per call at a 91% completion rate; 1,750+ automated tests in CI.
• CleverAnswerAI (US dental clinics): 100% of calls answered, +28% bookings compared with human operators. cleveranswerai.com
• Multilingual enterprise voice agents (EN, DE, FR, RO) integrated with CRM and ITSM systems.

Appointed expert

• Expert contributor to the Commission’s technical working group on AI economy policy; Month 7 deliverable cycle and April 2026 plenary.
• Core contribution, cited by a peer member (Bosch) as key: moving AI oversight from point-in-time audits to continuous governance with explicit accountability.
• Related: AI Chamber CEE Pro member (Legal Working Group, Prague Summit 2026); technical reviewer for Manning Publications (“Efficient AI”).

Staff AI Software Engineer

Led production AI across fraud, retention, compliance and support quality for a large iGaming operator, from research and system design to business outcomes.

• LLM-based QA for support: raised chat quality-control coverage from 2% to 25% with Schema-Guided Reasoning over millions of multilingual chats; assessment accuracy from 66% to 91%.
• Enterprise RAG platform on Node.js, AWS Kendra and Bedrock at terabyte scale; led pre-sales and solution design. Retrieval time from 15 minutes to 2–3.
• Real-time voice agent for marketing: 500+ concurrent calls on the OpenAI Realtime API with tool chaining, monitoring and fallbacks; saved thousands of agent-hours a month.
• Answering machine detection on SageMaker and Triton handling 1,000+ concurrent streams; cut wasted call costs by up to 60%.
• Fraud detection with behavioural biometrics: multi-accounting and bonus-abuse detection from mouse, typing and betting patterns that antidetect browsers and VPNs cannot hide. Target: 50–60% less bonus abuse.
• Uplift-based retention: bonuses targeted at persuadable players only, with LLM-personalised messages. Target: reactivation from 8% to 15–23% on 40–50% of the bonus budget.
• Responsible gambling (LLM-based): combined analysis of player messages and behaviour with an audit trail for regulators (UK, Sweden, Netherlands). Target: 200+ daily alerts down to about 25, false positives from 85% to about 12%.
• VIP early detection from behavioural signals instead of threshold rules. Target: 40–60% faster identification.

Solutions Architect, Team and Tech Lead

• Led the rebuild of the core LLM-agent platform: client bots built 40× faster, 80× lower cost, 10× faster deployment, 20× lower maintenance cost.
• Set up an engineering grading system and development standards for the team.

Master’s in Data Science, focused study period

Graduate coursework in LLMs, data science and MLOps.

Big Data Engineer, Machine Learning Expert

• Built an internal data warehouse from scratch with pipelines processing terabytes.

Python Software Engineer, Data Engineer

• Built a data warehouse from scratch for the data science team, enabling new user-facing products across the network.

Python Software Engineer, Data Engineer

• Built a data warehouse and data marts from scratch for data-driven decisions by divisional leadership.

Python Backend Developer, ML Developer