Langfuse Experts in Germany

• This project involved extraction of intelligence data to support report writing for a company that provides geopolitical, global, commercial intelligence. The data have been gathered from a number of resources (interview transcripts, online data, internal documents), and then a knowledge base has been build from it. This was the basis of a complex RAG system, that was evaluated against a golden dataset. Agents have been used to find out the contradicting intelligence, the statements supporting each other, and to store back the generated knowledge.

Used: Python, RAG, LangGraph, LangChain, deepeval, MCP

• Shipped Recare Voice Desktop from 0 to 1 in two months, including multi-language clinical documentation that auto-transcribes into structured German medical notes.
• Reduced LLM inference costs by 60–70% across Docs and Extract through prompt caching architecture.
• Built the AI workbench used by PMs/engineers for prompt experimentation and the Langfuse eval stack (10k+ traces evaluated).

Task:

• Design and build a proof of concept (PoC) for a future-proof virtualization platform, taking secure system architectures into account
• Assess the current state of existing infrastructures and develop selection and evaluation criteria for the right OS virtualization platform
• Carry out the requirements analysis and then create and prioritize tickets in the ticket system
• Complete and continuously update a tool evaluation matrix based on PoC results
• Support team knowledge building through clear documentation of the approach and results in Confluence
• Enterprise analysis of existing hardware (creating different BoMs)

Technologies: Vmware, Vmware Aria Operations, Osism, Canonical OpenStack, FishOs, Linux, Terraform, Ansible, Confluence, Alma

• Own full-stack development of an AI-native enterprise platform built on TypeScript, React, Vite, tRPC, Hono, and PostgreSQL, delivering AI-powered consulting workflows to B2B clients.
• Designed and shipped a multi-agent AI system using ReAct framework and Claude skills-style workflow patterns, including an intelligent PM assistant with rich system prompts, slash commands, tool integrations, and streaming chat UI.
• Architected an LLM evaluation framework: rubric-based LLM-as-judge, golden datasets, regression testing, and automated quality gating — ensuring consistent AI output quality at scale.
• Integrated LangFuse for end-to-end LLM tracing, conversation replays, and evaluation pipelines, enabling data-driven prompt optimisation that reduced token costs and response variance.
• Built with Drizzle ORM, pgvector, and knowledge graphs for structured data access, semantic search, and relationship-aware AI reasoning across the platform.
• Led TanStack React Query migration across the application — replacing manual state management with centralised caching and automatic refetching, reducing data-fetching boilerplate significantly.
• Practiced AI-native development throughout: Claude Code, Codex, Perplexity SDK, and LLM-assisted testing across the full development lifecycle. Deployed on Vercel + Azure ACA with Biome for linting/formatting.

Independent engineering practice focused on custom AI systems, production delivery, and fractional technical leadership.

Selected client engagement: Scalutions

• Role: Serve as fractional CTO and hands-on technical lead, responsible for the architecture and agentic infrastructure behind its managed B2B outbound operation.

• Product: Designed and built OutboundLoop, an agentic SDR operating system for research, qualification, personalized outreach, campaign management, human approvals, measurement, and continuous improvement.

• Scope: Own the full system lifecycle—from business processes and agent behavior to context design, model routing, integrations, evaluation, telemetry, reliability, cost control, and production operations.

• Led the end-to-end delivery of enterprise full-stack web applications by driving requirement analysis, solution design, frontend and backend development, database design, API integration, code reviews, team coordination, Agile execution, CI/CD deployments, production support, performance optimization, security implementation, and stakeholder collaboration to deliver scalable, high-quality software solutions.

• AI Integration (LLM & RAG): Design and build of an internal intelligent RAG system (Retrieval-Augmented Generation) based on LLMs, n8n, and vector data for the automated analysis of technical documents and error logs.
• Design & Implementation: Design of a robust RS-232/UART communication interface for an SBC-based embedded device to control medical shaker systems.
• Architecture & Protocol Design: Implementation of a highly maintainable software structure (OOP, SOLID) and definition of hardware-close, resilient communication protocols including multithreading and advanced error handling.
• Quality Assurance & DevOps: Test automation using xUnit, integration tests directly on the hardware target, and maintenance of technical documentation according to strict medical technology standards via Azure DevOps.

Label: C#, .NET, LLMs, RAG, n8n, RS-232, UART, Multithreading, async/await, xUnit, gRPC/protobuf, Blazor, MudBlazor, EF Core, Visual Studio 2026, Azure DevOps

• Converts natural language into JQL using RAG and pgvector. Returns structured JSON with a query, chart spec, and plain-text answer. A two-stage router answers general questions without touching the JQL pipeline at all.
• Interchangeable LLM backends: Ollama, vLLM, Groq, Anthropic Claude, AWS Bedrock - switchable at runtime, no code changes. Self-healing JQL: on Jira validation failure, feeds error back to LLM, retries up to 4 times. OCI Vault for secrets. Deployed on Oracle Cloud A1 with GPU inference over Tailscale private network. Open source.

End-to-end responsibility for all products - concept, architecture, development, and production operation as the sole developer; in addition, customer meetings, proposals, and marketing.

Underwriting Copilot - AI assistant for industrial insurance (in production at customer sites)

• Supports underwriters in analyzing industrial insurance submissions - in production use at an industrial insurer.
• Framework-independent RAG architecture with Hybrid Search (BM25 + pgvector) across large, mixed document sets.
• Two-stage evaluation and observability pipeline (code assertions + LLM-as-Judge) that makes answer quality, retrieval accuracy, and citation integrity measurable in a regression-safe way.

Kaiki Menu Analyzer - Data intelligence platform (in production at customer sites)

• Automatically captures and analyzes menu data from around 25,000 German restaurants.
• Scalable 7-container architecture (FastAPI, partitioned PostgreSQL, Redis/RQ) with LLM-supported extraction of structured data from PDF, HTML, and images.
• Full CI/CD pipelines (GitHub Actions), production cloud deployment, interactive dashboards (Dash).

Kaiki GEO Atlas - GEO platform (in production at customer sites)

• Measures brand visibility across five AI engines (ChatGPT, Gemini, Perplexity, Grok, Claude), each augmented with web search, orchestrated as a DAG workflow pipeline (Dispatcher → Sub-workflows → Scoring → Report) with fail isolation.
• 6-container deployment (FastAPI, Celery, Redis, PostgreSQL); LLM cost estimation, PDF audit report, rule-based cross-signal insights (no extra LLM cost).

Data Pipeline & Analytics Platform - competitive analysis in the automotive aftermarket

• Automated data pipeline with gap analysis algorithms and role-based access control; 230+ tests.
• Backend with FastAPI, PostgreSQL, SQLAlchemy.

Product development (actively in progress)

BankingGPT - AI assistant for complaint management in cooperative banking

• Security architecture at the core: no AI draft reaches the customer without human approval - the approval decision is in auditable code, not in the language model (monotonic: the model may escalate, never downgrade).
• Real agentic building blocks, each with its own boundary: the model chooses tools itself through an MCP server (read-only, allowlist, capped, fail-safe); sensitive cases are handed off via an open A2A protocol (JSON-RPC, Agent Card, message/send/tasks/get; client implemented by me) to a separate specialist agent (securities/law), which never lowers the review requirement (pinned by test).
• Evaluation-driven over ten analysis rounds; uncovered a security flaw through independent review and blind tests that nine automated runs had missed.
• Voice AI frontend, responding live: covered cases are answered in the conversation, sensitive ones escalate before generation; response latency < 7 s measured (local GPU STT/TTS).

Stack & production readiness: Python, pydantic-ai, FastAPI/Celery, PostgreSQL/pgvector, FastMCP, fasta2a, Docker; multi-tenant capable (physical vector isolation per tenant), PII encrypted, OWASP-LLM reviewed, 275 tests, CI/CD; vendor-portable (Ollama / EU Cloud Vertex).

After-Sales Assistant - agentic RAG/GraphRAG assistant on public OEM manuals (automotive after-sales)

• Genuinely agentic on LangGraph: ReAct agent with four tools and conversation memory - the model decides on its own whether to use the manual (RAG, Chroma), a knowledge graph (GraphRAG, Neo4j/Cypher - decodes warning lights), or a workshop/booking service.
• Human-in-the-Loop before the irreversible action: before every appointment booking, the graph pauses (interrupt) and gets the driver's explicit confirmation - the same approval-before-action discipline as in BankingGPT, in a different framework.
• Eval as CI gate: a three-part scorecard (RAGAS grounding + deterministic tool-routing accuracy + DeepEval safety: does the answer mention the warning first when there is a critical warning?) blocks the pipeline; provider-agnostic (OpenAI/Azure/Anthropic), FastAPI with token streaming.

Stack: Python, LangChain/LangGraph, Chroma, Neo4j, RAGAS/DeepEval, FastAPI, Docker.

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

• Industry: B2B
• Headless frontend with AEM integration
• Key challenge: Migrating a PWA application in live operation based on .NET and legacy code; the entire application must be converted to React and Express.js/TypeScript
• Technical frameworks: Tailwind, XML, JavaScript, Caddy, ReactJS, Express.js, REST API, JSON
• Cloudflare CDN
• Caddy server with GitHub CI/CD pipeline

• Lead a cross-functional squad of backend, frontend, and ML/data engineers, balancing hands-on contribution (architecture, coding, reviews) with team leadership (mentoring, backlog prioritization, roadmap alignment).
• Designed and delivered ML-powered search and discovery features, including Learning-to-Rank (LTR), query expansion, and vector search, improving result relevance and user engagement.
• Implemented personalization and recommendation pipelines, using behavioral data and segmentation to increase customer retention and lifetime value.
• Established data-driven practices, building A/B testing and experimentation workflows (Odyn, MLflow) to measure feature impact on CTR, NDCG, and conversion.
• Owned the squad’s architecture and delivery roadmap, modernizing services with cloud-native microservices and event-driven systems (AWS, Pulumi, Terraform) to improve scalability and reliability.
• Improved reliability and operational excellence, introducing observability (Prometheus, Grafana, NewRelic), incident management, and postmortems that reduced downtime for customer-facing services.
• Mentored and supported engineers, fostering technical growth, collaboration, and a customer-first mindset through regular feedback, coaching, and code reviews.
• Worked closely with product managers, researchers, and business stakeholders to translate customer insights into technical solutions that improved discovery, engagement, and retention.
• Explored Generative AI/LLM use cases (GPT-4, LangChain, RAG), prototyping intelligent assistants and personalized discovery workflows that increased user satisfaction.
• Delivered tangible results: boosted engagement through personalization, contributed to revenue uplift, and reduced incidents by embedding resilience and observability.

• Conceived, built and iterated an applied-AI MVP that turns in-depth audio interviews into structured, long-form narrative outputs across multiple genres (e.g. memoir, institutional knowledge, thematic essays) using agentic orchestration and multi-step reasoning.
• Designed and implemented core workflows in a Next.js-based stack, working with structured representations (JSON and other formats), retrieval-augmented generation and emerging knowledge graph structures to maintain context and consistency over long documents.
• Defined and tested agent behaviors across realistic storytelling scenarios, including ideal user journeys, edge cases and failure modes, with explicit criteria for coherence, factual alignment and user intent satisfaction.
• Currently running targeted user tests with selected partners to validate use cases and inform the next product iterations.

• Designed and implemented multi-modal entertainment toolchains that combine passenger input, vehicle context, large-language models (text-to-text and speech-to-speech) and image generation models to deliver more interactive and immersive in-car experiences.
• Built and orchestrated tools for LLM-based agents, covering session management, background task execution, dynamic user interactions and persistent application state.
• Investigated multi-agent orchestration frameworks for in-car environments, evaluating communication protocols and architectural strategies for coordinated and reliable agent behavior.