Generative AI Engineers in Berlin

• Analyze the root causes of complex functional, technical, and data-related errors in PowerCloud across process, booking, and system boundaries.
• Identify automation and straight-through processing opportunities at the interface between business teams, IT, and external service providers.

Built and scaled a high-performance enterprise API for real-time address validation and data quality with sub-second latency and 99.9 % availability.

Designed and integrated the solution into e-commerce, checkout, and logistics processes of leading European companies. Reduced delivery errors and shipping costs through automated data correction and precise data validation.

End-to-end responsibility for product strategy, technical architecture, software development, enterprise customers, operations, and GDPR-compliant data processing. Combined AI-native engineering workflows, Python, SQL, API integration, data quality, and workflow automation.

The MiDAS program aims to integrate Generative AI across the entire BBM organization. My role is to enable the eBike division to adopt and leverage Generative AI by collaborating closely with the MiDAS team in India.

Management:

• Enable the eBike division to adopt GenAI through tailored solutions.
• Collaborate with the MiDAS team in India to design solutions for Generative AI use cases specific to the eBike division.
• Coordinate with the eBike teams to integrate GenAI systems, tools, plugins, and services provided by the MiDAS team.
• Plan roadmaps with eBike and MiDAS India team Product Managers for the delivery and integration of Generative AI solutions.

• ▶ Built real-time voice assistant (STT → LLM → TTS pipeline) benchmarking and evaluating multiple STT providers including faster-whisper and Azure Speech. achieved sub-3s latency, Groq API (Llama 3) with multi-turn memory - directly handling edge cases in dictation, names and passcode recognition.

• Designed and deployed enterprise Retrieval-Augmented Generation (RAG) applications using LangChain, LangGraph, vector databases, embeddings, and open-source LLMs served through vLLM on GCP GPU infrastructure.
• Built agentic AI workflows using LangGraph with planning, reasoning, tool execution, persistent memory, session management, and Human-in-the-Loop approval mechanisms.
• Developed LLM-powered automation systems integrating BigQuery, SQL pipelines, and external advertising APIs including Meta, TikTok, Amazon, Snapchat, Google, and Pinterest, reducing manual operational workflows.
• Architected multi-agent AI systems for enterprise analytics and decision-support workflows, enabling autonomous task execution and intelligent data interactions.
• Implemented retrieval optimization strategies including multi-retriever architectures, semantic search, context optimization, and query improvement techniques, improving response relevance by approximately 40%.
• Engineered structured prompting strategies, function-calling schemas, and validation workflows to improve reliability of multi-step LLM applications.
• Designed scalable AI services using Python, FastAPI, Cloud Run, Pub/Sub, BigQuery, Docker, and cloud-native deployment architectures.

• Building an AI-powered (LLM-based) application to take the chaos out of job hunting using GenAI tools like Lovable & Gemini
• Currently in alpha with a set of 10 users to capture user feedback for early iteration on the features, UX and LLM prompts

CRM and revenue platform managing billing and subscription workflows.

• Scaled payments infrastructure across data products, direct debit expansion and automated abuse prevention, generating $416K in annualized operational savings ($8K/week) by eliminating redundant gateway calls.
• Owned backlog and sprint execution for autonomous checkout abuse detection pipelines, designing real-time risk guardrails and velocity heuristics that blocked card testing attacks.
• Architected enterprise billing migrator user stories and data reconciliation mechanisms, achieving zero-downtime subscription state transitions and cutting $60K in infrastructure overhead.
• Expanded European direct debit (SEPA) payment capabilities, managing cross-squad API dependencies and automated webhook error-handling to eliminate checkout friction.

Delivering two large, complex strategic initiatives in parallel with an 8-person team – OCFA (platform migration and consolidation of the cart-checkout backend) and Conrad PRO (a new paid loyalty/membership programme) – with agentic AI development as the team's standard way of working rather than an experiment on the side. Set technical direction, engineering standards and quality gates across both, hands-on in the codebase throughout.

Agentic development as the delivery model:

Introduced LLM-based development (Claude Code, Gemini) as the core team workflow and built the agent tooling behind it – custom Claude Code skills encoding the team's coding standards so generated code arrives conformant, multi-agent sub-agents for code review and legacy-parity checking, and a scheduled diff-monitoring loop watching the legacy system for changes. CI build time dropped by ~36%, and the team sustained roughly two production changes per day: 38 parity and feature tickets closed via 39 reviewed and merged MRs in a representative four-week window.

Quality ownership over AI output:

Every agent-generated change passed the same gates as any other – automated tests, parity validation and SonarQube – with human review before merge. Speed came from the agents; accountability for what shipped stayed with the team.

Java backend & legacy-parity migration:

Java 25 and Spring Boot 4 with Spring Modulith, structured around hexagonal architecture and DDD. Defined the shadow-testing strategy for legacy-parity correctness – reverse-engineering undocumented legacy contracts and validating against real production traffic – through to shadow-parity readiness for the first live country cut-over (Netherlands).

Enterprise integration:

Integrated downstream systems (SAP, payment, pricing/offer, shop-config) via contract-first OpenAPI and outbound adapters, using Kafka and Pub/Sub for asynchronous messaging.

Stack: Java 25, Spring Boot 4, Spring Modulith, Hexagonal Architecture / DDD, Contract-First OpenAPI, Kafka, GCP Pub/Sub, SAP integration, PostgreSQL, GitLab CI/CD, SonarQube, Claude Code, Gemini

• Built a hybrid semantic and keyword search and LLM-based requirement extraction from conversational queries, boosting search accuracy by 85%, cutting zero-result searches by 70%, and reducing search time by 60%.
• Deployed a production-ready API with monitoring dashboards over 100K+ products, keeping response times under 2s and reducing customer search-to-purchase time by 40%.
• Technologies: Python, BGE-M3, Qwen2.5, FastAPI, Qdrant, Meilisearch, Docker, Prometheus, vLLM.

Enterprise AI content categorization and AI-powered web research.

• Built multi-LLM evaluation framework with annotated data
• Iterated LLM error rates based on annotated datasets
• Implemented AI-powered web research pipeline Stack: LLM, evals, OpenRouter, Python, Node.js, TypeScript, React