Vlad B.

Brands owned by Rollins, the S&P 500 company (~$3.4B revenue; NYSE: ROL) are currently using AI sales agents I built. One sold 769 accounts and generated $620K for a single customer in a year (public case study on sellifyai's website); another made $1M in a month. From CEO Thomas K. Lundberg ($350M exit): "Vlad has been incredible to work with" If you shoot me an invitation or message i'll send you a personalised video back on how i may be able to help you! A little more about me: 👋 Hi! I'm Vlad - Senior AI Engineer · Top Rated Plus (top 3%) · Claude Certified Architect, Anthropic. Agentic AI and production LLM applications, AI automation. 🔧 THREE THINGS I DO 1️⃣ Agentic AI & AI automation in production. Multi-agent systems, agentic workflows and agent orchestration (LangGraph, Claude Agent SDK, CrewAI, LlamaIndex) with tool calling and structured outputs, wired into your CRM and internal tools: HubSpot, Monday com, SharePoint, Microsoft 365, and the legacy CRMs. I help you pick what stays deterministic code and what needs an agent. Sensitive steps wait for human approval, every action lands in an audit trail, and an evaluation framework runs on every change so quality holds at scale. I also take over half-built agent projects that broke with real users and carry them from prototype to production. 2️⃣ Claude ecosystem engineering. Claude API, Claude Code, Claude Agent SDK, Anthropic API, Claude Cowork, Claude Skills. MCP server design (Model Context Protocol), prompt caching, model routing, token cost optimization. I have cut Anthropic API bills, with before/after evals proving quality held. I also set teams up with AI coding agents: Claude Code and Cursor in the daily loop, code review and pull requests still owned by people. 3️⃣ Private LLMs, self-hosted. Open-weight models (Qwen, Llama, Mistral) with vLLM model serving on your own GPU, Ollama for local runs, LiteLLM gateways, fine-tuning when a prompt will not get there. Builds stay provider-neutral, so a hosted model can move on-premises later without rebuilding retrieval. GDPR and data residency come up constantly with EU clients. ✅ HOW I WORK Production AI systems break once the prompt has been edited twenty times. I prevent this with: • Evaluation frameworks, automated testing and regression tests on every change; model evaluation protects against regressions. • Guardrails, prompt injection defense and human-in-the-loop review. Answers stay grounded in your own sources, so hallucinations get caught. • Reliable AI plumbing: retries, idempotency, queues, audit logging, observability, monitoring. Least-privilege access and secrets management, so agents never touch raw credentials. RBAC and multi-tenant setups. • Cloud deployment your team can maintain (Docker, AWS, GitHub Actions CI/CD), with docs and training. 📈 RESULTS (public case studies) Two years on the founding team at Sellify AI, outcome-priced AI sales agents for US pest control, paid only when the AI sells: • Over $1M in new revenue in one month for HomeTeam, one of the largest US pest control firms • $620K in new sales and 1,100 new accounts for Ninefold • I built the legacy CRM integrations (FieldRoutes, PestPac) rivals still have not copied. They closed the six-figure clients Since then, contract Senior AI Engineer at a recruitment SaaS, working directly with the founder: a job-monitoring agent live at several agencies with a human approval step, plus an evaluation framework (Braintrust, Promptfoo, also custom evals). Also a legal RAG agent over an 89,600-section Norwegian law corpus, quote-verified against the law. ⭐⭐⭐⭐⭐ "Very sharp and understands the intricacies and needs of our company. Have trusted him with important tasks, and he's always delivered." - Thomas K. Lundberg, CEO Sellify AI ($350M prior exit) 🚀 HOW TO START 🔍 Free teardown, 48 hours. Send 10 traces where your AI got it wrong. You get the three root causes and what I would fix first, in writing. 📋 AI Assessment & Roadmap, from $2,500, ~2 weeks. Architecture, what it costs at your size, what breaks first, a plan in priority order. 🛠️ Then we build. I work independently. Milestone or hourly, EU/US hours. 🛠️ STACK AI/LLM: Anthropic Claude Code, Claude Agent SDK, OpenAI, Azure OpenAI (EU/GDPR), AWS Bedrock, Gemini · multi-agent and LLM orchestration (LangGraph, LangChain, CrewAI, LlamaIndex) · MCP server design · RAG and AI knowledge bases (pgvector, Qdrant, hybrid search, reranking, embeddings) · LLMOps: evals, observability, prompt caching, model routing, cost optimization · generative AI (GenAI), guardrails, structured outputs, tool use / function calling · fine-tuning, vLLM, self-hosted open-weight models (Qwen, Llama, Mistral) · document extraction, Jev, OCR, AI workflow automation Engineering: Python, TypeScript/Node.js, PostgreSQL, Redis, Celery, Docker, AWS, GitHub Actions CI/CD, REST/webhooks/OAuth · Shopify, Slack, Notion, WhatsApp Business API Domains: vertical SaaS · recruitment