pgvector Experts in Germany

• TrendReel, production LLM agent and RAG system (Python, LangChain, OpenAI, Groq/Llama 3, Claude, FastAPI, Kubernetes, PostgreSQL).
• Designed and built a production multi-step LLM agent system: a script generation agent with a per-platform psychology database, 7 viral narrative frameworks, and structured quality scoring, switching between Claude and Groq backends in real time based on output metrics.
• Implemented multi-provider LLM routing (Claude primary, Groq/Llama 3 fallback) with priority-chain failover and quality-based provider switching, achieving 95% inference cost reduction while holding measurable quality thresholds.
• Built an advanced RAG-style retrieval pipeline with per-platform knowledge bases, semantic content matching, and structured output evaluation across 7 decision frameworks, directly analogous to multi-tenant context-based reasoning for enterprise document workflows.
• BrainyAI, adaptive AI learning platform (Python, LangChain, Groq Llama 3.3-70B, OpenAI, Next.js, Supabase, Redis).
• Integrated Groq Llama 3.3-70B with education-level-aware prompting, dynamically adjusting vocabulary depth, citation complexity, and reasoning style across four student proficiency tiers.
• Nexus Prime, multi-tenant SaaS platform for marketing and growth automation (25 modules, 99 backend routers, 153 frontend files).
• Built a 25-module, 99-router multi-tenant SaaS platform covering ad remix, affiliates, WhatsApp inbox, email, and cart recovery, serving four subscription tiers from $199 to $1,999 per month with integrated Stripe, Paystack, and Flutterwave billing.
• AI Video Surveillance Platform, multi-tenant edge and cloud computer vision system currently in active client pitch.
• Designed a multi-tenant AI video surveillance platform combining edge YOLO26 inference on NVIDIA Jetson Orin NX boxes with a central GKE cloud layer (Postgres, Pub/Sub, ClickHouse, R2, Keycloak) for event storage, dashboards, alerting, and multi-tenancy.

• 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.

Development of an iOS app for virtual try-on and outfit suggestions

The goal of the project is to develop a mobile application for personalized, photorealistic outfit suggestions. Users should be able to upload their own photos, try on clothes virtually, and find products that can be bought directly in the generated suggestions.

• Planning and implementation of the onboarding and photo upload in the iOS app
• Development of the mobile application with Expo and React Native
• Implementation of a Hono/Node.js backend for user, product, and generation processes
• Building an asynchronous processing pipeline with BullMQ and Redis
• Connection of PostgreSQL/pgvector and S3 for product, image, and generation data
• Integration of Gemini and OpenAI for outfit generation and image processing
• Implementation of a credit system and integration of RevenueCat
• Integration of Stripe Connect and affiliate product feeds for products that can be bought directly

Label: TypeScript, React Native, Expo, Hono, Node.js, PostgreSQL/pgvector, BullMQ, Redis, S3, Gemini, OpenAI, RevenueCat, Stripe Connect, Docker

Manufacturing intelligence platform with enterprise workflows, RAG, real-time AI assistant features, and multi-repository backend architecture.

• Built and extended production AI/backend services with Django, DRF, FastAPI, GraphQL, Celery, PostgreSQL, MySQL, Redis, and WebSockets across a modular multi-repository platform.
• Contributed to ARIS V2, a real-time manufacturing AI assistant using LangChain, LangGraph, MCP tool orchestration, planning/execution flows, OpenAI, AWS Bedrock, Qdrant, and Elasticsearch/OpenSearch-backed retrieval.
• Supported rollout expansion from ARIS V1 in 4 of 17 client production plants to ARIS V2 currently active in 13 of 17 plants, increasing real-world deployment coverage to more than 50% of the client footprint.
• Worked on document-grounded RAG functionality including ingestion, OCR, chunking, embeddings, indexing, retrieval, reranking, and grounded answer generation for industrial workflows.

Stack: Python, Django, DRF, FastAPI, LangChain, LangGraph, GraphQL, Celery, WebSockets, OpenAI, AWS Bedrock, Qdrant, Elasticsearch/OpenSearch, PostgreSQL, MySQL, Redis, Docker.

• Mentor team of 6 ML platform engineers through weekly 1:1s, technical design reviews, and best practices, improving team velocity by 35% through structured sprint planning and skill development programs
• Define 2025–2026 ML platform roadmap in collaboration with Data Science, Cloud Engineering, and Product teams, prioritizing automated model governance, cost attribution systems, and multi-environment deployment strategies
• Partner with Data Science, SRE, and Product stakeholders to align ML platform capabilities with business objectives, reducing data scientist deployment friction by 60% through self-service platforms
• Architect and deliver production-grade MLOps platform supporting 50+ models in production with automated promotion pipelines, versioning, and rollback capabilities, achieving 99.5% platform uptime SLA
• Design distributed ML pipeline architecture using Metaflow and Argo Workflows (Vertex Pipelines-compatible), reducing model training time by 30% and deployment cycles from 2 weeks to 3 days through full CI/CD automation
• Build containerized ML services on Kubernetes with auto-scaling policies, resource quotas, and multi-tenancy isolation, optimizing infrastructure costs by $180K annually (25% reduction)
• Implement monitoring, alerting, and performance tracking using Prometheus, Grafana, and custom instrumentation, reducing model debugging time by 50% and establishing model performance SLOs
• Lead development of RAG-based document intelligence platform using LangChain, LangGraph, and vector databases, implementing agentic AI workflows for automated financial document processing
• Implement Infrastructure-as-Code using Terraform for reproducible environment provisioning and GitOps workflows, reducing infrastructure drift incidents by 80%
• Design role-based access control for ML platform, implement model lineage tracking, and establish audit trails for regulatory compliance aligned with enterprise IAM best practices

Technologies: n8n, Notion, OpenAI API, Claude, MS AI Foundry, MS CoPilot Studio, MS CoPilot, LLM, Node.js, Vercel, LangGraph, PostgreSQL, pgEdge, pgvector, Docker, LangChain, Ollama, Open WebUI

• Continuous evaluation and prioritization of internal automation needs
• ~20 AI agents in active use: research, content pipelines, document processing
• 5 n8n workflows for automated data and process control
• Architecture built on the same principles as in customer projects: state management, event-driven orchestration, API integration
• Ongoing operation and further development

AI Platform Backend – Senior Developer

Brought in to design and build a backend for an AI platform from scratch, including multi-provider LLM orchestration and real-time infrastructure for AI influencer personas at scale.

Tasks and responsibilities

• Architected and implemented a multi-LLM orchestration layer with Semantic Kernel to integrate GPT-4 and other providers for core platform logic and AI influencer personas, reducing model-switching overhead by abstracting provider APIs behind a single interface.
• Designed and developed a backend from scratch in C# / .NET 10, including domain modeling with DDD, a versioned RESTful API layer, and cloud infrastructure setup on Azure.
• Built a real-time chat infrastructure with Server-Sent Events (SSE), message persistence, and delivery guarantees for live operation of AI influencer personas at scale.
• Developed a media management service with integration of cloud object storage for upload and retrieval of influencer-generated content.
• Created an integration and unit test suite with data seeding for reliable regression testing across all core platform flows, significantly reducing the production error rate.

Tools and technologies: C#, .NET, ASP.NET Core, Python, TypeScript, MySQL, Semantic Kernel, EF Core, Minimal APIs, LLM Orchestration, Prompt Engineering, Agentic AI, Generative AI, AI-Assisted Engineering, Claude Code, GitHub Copilot, Google Gemini, OpenAI API, Ollama, Redis, Azure, Azure Container Apps, Azure Database for MySQL, Docker, GitHub Actions, Clean Architecture, Vertical Slice Architecture, CQRS, Domain-Driven Design, REST API, xUnit, Integration Testing, Unit Testing, Jira, Confluence, Scrum

• 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.

• Built a JTBD-based prioritization framework for 3,000+ accumulated feature requests, identified 27 broker jobs, validated 8 through 25 user interviews, and used the resulting job map as a live prioritization filter for all incoming channels (Upvoty, CSAT, consulting tickets).
• Responsible for the Scout24 Lighthouse initiative: Document Intelligence with full RAG architecture (semantic chunking, bge-m3 embeddings, pgvector, BM25+Dense hybrid retrieval).
• Reduced lead time of customer feature requests to 3.1 days through code analysis, ticket specification, and independent implementation using a coding agent (Codex).
• Developed an LLM-based support agent (GPT-4o mini, Codex-generated merge requests) that reduced 3rd-level escalations from 40% to 5% of all monthly tickets.
• Integrated six partners through technical coordination, specification, backlog and release management, and led seven full stack developers.
• Eliminated regulatory exposure for brokers in six weeks through risk analysis (BGH ruling on distance selling/GDPR), new audit features, and coordination with legal and data protection officers.

• Delivered a new notifications platform implementation built from scratch to replace existing and upcoming services
• Collaborating with other teams to integrate more domains

Tech stack:

• Data: Scala 3, Apache Kafka, Python, Airflow, Astronomer
• BE-FE: TypeScript, NestJS, Java, Spring Boot, Vue
• Dev-ops: AWS, PostgreSQL, Docker, GitHub Actions, Kubernetes, Helm, ArgoCD

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

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.

• Developing a professional scheduler for organizations with specific detailed requirements
• Evaluating different existing software solutions
• Creating a list of technical requirements
• Implementing these requirements
• Selected technologies: WebDAV, CalDAV, Rust, Baikal, OAuth, Keycloak

• Insurance service provider*

Challenge: Business-critical documents were processed through external OCR providers, with ongoing costs, dependency, and data privacy risks for sensitive insurance data.

Implementation:

• Architecture and production implementation of an on-premise OCR solution with full data ownership
• Methods for recognizing document structures as the basis for automated further processing
• ML-, NLP-, and LLM/VLM-based information extraction, especially from invoices and quotations

Success: Replaced external providers: full data ownership, GDPR-compliant processing, and 75% lower recurring OCR costs per year

Used technologies: Python, Docker, Microservices, FastAPI, PyTorch, Torchvision, MongoDB, MySQL