Ramat Gan, Tel Aviv District, Israel
Applied AI/LLM Engineer who ships production-grade LLM systems end-to-end - agent architectures, RAG, MCP context engines, evals and guardrails - that change how engineering teams build, test, and ship software. I've been building with LLMs since 2023, before the agentic-coding wave. Today I own production AI features in a cybersecurity product, MCP-based dev-tooling used by other engineers, an AI-First SDLC strategy, and the model platform serving 150+ engineers across a 600-person company. What that looks like in practice: - Ship production AI features - a conversational form-filling assistant taken from prototype to production solo; a context engine for coding agents; automated SBOM license/copyright enrichment raised from ~40% to 98% accuracy and gated in CI/CD. - Run the model platform - deploy and route models behind a unified API (Azure AI Foundry, LiteLLM, multi-region) with cost/usage tracking. - Drive AI adoption org-wide - governance for agentic coding tools, end-to-end tool trials (Copilot / Cursor / Claude Code / Codex), a Center of Excellence, and talks at AI-engineering conferences. I prototype fast, iterate with real users, and productize. Data-engineering and signal-processing background, with a track record of ramping quickly on unfamiliar domains (currently: chip-design verification and formal methods). --- "We're only as smart as evolution has made us so far…" - I'd rather manage the machines than be replaced by them. If we don't want to be replaced by AI, we'd better start acting like it.
Applied AI/LLM Engineer shipping production LLM systems into Semperis's cybersecurity products and engineering workflows, and driving AI adoption across the org. Production AI features: - Own Spark, Semperis's first production AI feature: a conversational assistant for a complex form that completes it from the user's inputs or guides them through it. Took it from prototype to production solo, coordinating Backend, Frontend & UX/UI with no formal resource mandate. - Built Glyph, an MCP-based context engine managing AI memory/context for coding agents across large codebases, adopted by other engineers. - Owned SBOM license/copyright enrichment quality end-to-end: built a labeled benchmark, ran controlled experiments across five approaches, and raised accuracy from ~40% to 98% via LLM-assisted verification, then gated the metric in CI/CD to block regressions. Model platform & ops: - Set up and operate the company's shared model platform on Azure AI Foundry, deploy and route models behind a unified API (LiteLLM, multi-region) with cost/usage tracking, serving 150+ engineers. AI enablement & governance: - Designed an AI-First SDLC strategy mapping agent insertion points across spec → architecture → code → QA → deploy → monitor. - Authored a three-layer governance policy for agentic coding tools; ran security review of submitted Skills, MCPs & Plugins. - Led 4 concurrent AI-coding-assistant tool trials (Copilot, Cursor, Claude Code, Codex) end-to-end, ending in a data-driven adoption recommendation. - Run AI enablement org-wide (CoE wiki, AI champions network, weekly session) and speak at external conferences.
Data Engineer with an early AI-engineering bent - built one of the company's first LLM-powered internal tools in 2023, alongside core data-pipeline work. - Built an in-house, self-correcting LLM code-generation loop (generate → run → fix → regenerate) wired to internal systems - an early, small-scale agentic coding assistant in 2023, ahead of the agentic-coding wave; helped spec the company's Generative-AI direction. - Rewrote a critical legacy nightly process end-to-end, cutting runtime from 5 hours to 3 minutes (~95% reduction) and saving ~35 hours of processing per week. - Implemented parallel processing across thousands of daily API calls, drastically improving throughput and reliability. - Closed long-standing security gaps in legacy code and stood up a logging & alerting system that made incidents detectable instead of silent. - Refactored toward a modular structure that set a new internal bar for maintainability and security. - Partnered with stakeholders through deployment, turning a fragile nightly job into a reliable, faster data operation.
• Developed a real-time dashboard for the Investments & Risk department, tracking 20+ cryptocurrencies and providing critical insights. • Optimized data pipelines using Python and SQL for enhanced performance and reliability. • Analyzed blockchain data, enabling informed decision-making within the department. Note: The department was disbanded following the 2022 FTX collapse.
• Developed and managed analytical products for hedge funds, leveraging user data to provide actionable market insights. • Analyzed usage patterns across the US and Europe using Python, SQL, Excel, and Google Sheets, helping hedge funds gain a competitive edge before earnings calls. • Collaborated with engineers and scientists to design data pipelines and develop analytical models, enhancing data accuracy and relevance. • Created and delivered dynamic Excel-based products, presenting complex data analysis in a user-friendly format to clients.
• Develop and maintain Spetz's Chrome Extension (Currently "Paradox").