Martin EnkeFull-stack & AI Engineer

To better understand the Java ecosystem, I built a small full-stack incident management tool using Spring Boot and Angular — technologies that were largely new to me. The goal was to learn how these frameworks structure applications by implementing a simple but complete system end-to-end. The backend exposes REST endpoints for managing incidents, including validation and filtering, while the Angular frontend provides a lightweight interface for creating and reviewing tickets. Building the system from scratch required working through typical concerns such as request handling, cross-origin communication, and how frontend components interact with backend services. What made the exercise particularly valuable was seeing how common architectural patterns appear across different stacks. Even in a small project, separating concerns between API layers, service logic, and UI components helps keep the system understandable and easier to extend. Exploring these patterns in an unfamiliar environment reinforced how similar design principles apply regardless of the specific framework. Projects like this complement my work in AI systems by strengthening the underlying application architecture that intelligent services ultimately rely on.

As the internship progressed, my focus shifted from onboarding into contributing across different parts of the system. Daily standups and ongoing collaboration helped establish a shared understanding of priorities and system state. I started working on concrete tasks within a layered service structure, gaining hands-on experience with how specifications, application logic, and infrastructure concerns are kept deliberately separated. A recurring theme has been learning how small changes propagate through well-defined layers, and how clear boundaries make it easier to reason about behavior, testing, and future extensions. Seeing these patterns applied consistently in a real-world codebase has been especially valuable.

Started building an Agentic AI Operations Console as a high-stakes operations blueprint for supervising agentic AI workflows in support, compliance, trust & safety, and internal operations. Core design principles: • Human-in-the-loop by default — AI assists, humans decide • Explicit agent handoffs with full visibility into status, priority, and escalation context • AI-generated artifacts (summaries, risk signals, reply drafts) are read-only, versioned, and auditable • Failure-safe design: AI errors never block core operational workflows The console is intentionally not a chatbot frontend, but a decision-support system focused on transparency, control, and accountability.

Visited the PyData Berlin event, which brought together practitioners and researchers across data engineering, machine learning, and applied AI. The talks and discussions reinforced how Python remains a central ecosystem for building, evaluating, and operating production-grade AI systems.

Started my internship at Webeet. The onboarding phase focused on understanding the system landscape, team workflows, and architectural foundations. A central topic early on was Hexagonal Architecture (ports & adapters) and how it is used in practice to separate domain logic from infrastructure, delivery mechanisms, and external services. Rather than diving straight into feature development, the initial weeks emphasized reading existing code, understanding boundaries between layers, and learning how architectural decisions support long-term maintainability in a production environment.

Prepared for my internship at Webeat by getting familiar with tooling, reviewing code standards, and setting up a small local practice environment to hit the ground running.

Started the Masterschool career preparation track: refining projects, polishing documentation, upgrading my portfolio, and preparing for the next step. A very motivating phase.

Explored the Model Context Protocol (MCP) and Berkeley Modular Autonomous Design (BMAD). Both offer compelling perspectives on agent orchestration, tool use, and structured agentic workflows. Lots of ideas brewing.

Finished the final Backend & GenAI Engineering exam with a score of 96.7%. Feeling proud of the journey and grateful for the progress over the past year.

Preparing for the final Masterschool exam in Backend & GenAI Engineering. Feels like the culmination of months of growth and countless small breakthroughs.

Expanded CV PDF ingestion with structure-aware parsing. Added bias-neutral scoring and introduced an ethics transparency dashboard to surface scoring logic and audit signals.

Attended the first Global AI Community “AI Show & Tell” in Berlin. Real demos, deep technical talks, and strong discussions around agent systems, evaluation, compliance, and AI memory. Inspiring energy — Berlin’s AI ecosystem feels alive and collaborative.

Built local, evidence-bound candidate scoring with bias auditing: LLM scoring with heuristic fallback, rubric-based metrics with quoted spans, blinding and prestige-proxy detection, and an evaluation dashboard (Spearman ρ, Top-K overlap, mean |Δ|). Added recruiter, ethics, and developer viewer modes with exportable audit trails.

Refined data schemas for phrase patterns and progression trees. Added tempo-linked notation syntax to explore how lesson content could adapt dynamically. A space where composing meets coding.

Started notation experiments for musical logic and rhythm training. Built minimal Python tools to map rhythmic ratios and interval structures as lesson scaffolding.

Kicked off Hirethics AI: FastAPI + Pydantic backend with rubric templates, anonymized inputs, initial LLM scoring loop, and heuristic fallbacks. Frontend scaffolding in React/Vite with clean viewer modes.

Implemented A/B split view with a synced timeline, sparklines for reward and error metrics, and an “Explain last 10s” LLM summary panel. Polished UI with scientific-style cards.

Split the project into orbital-rl and orbital-frontend. Built a clean Gymnasium environment, random-policy baseline, PPO training path, and first rollout analytics. Early separation between Random and PPO achieved.

Integrated the AI Promo Agent into my portfolio with hover previews and a scrollable blog section.

Final mobile header fixes, disclaimer overlay, favicon, and deployment on Vercel.

Built a creative pipeline from brief to palette to content. Implemented prompt composition, clean PDF export, and UX guardrails.

Added swing, metronome, and haptics. Sessions now persist via localStorage.

Built a mobile-first Web Audio step sequencer with a focus on timing accuracy and a friendly save flow.

Presented ZoundZcope live and received great feedback on the A/B compare workflow and RAG usefulness.

Prepared storyline, demo scripts, README updates, and short screen captures. Dry-runs with peers.

Improved logging, added rate limiting, and tuned cold-start behavior on Render.

Added loading states, error boundaries, and tightened layout to reduce time-to-first-insight.

Chunked documentation, built a retrieval pipeline, and grounded AI critiques with relevant snippets to reduce hallucinations.

Implemented side-by-side AI feedback across multiple versions with session save and restore.

Built context-aware follow-ups per upload with token management and safe truncation logic.

Launched the blog with multilingual support, AI moderation, translation, and read-aloud. First successful production deployment.

Integrated structured JSON critiques via OpenAI with caching, issue tagging, and actionable feedback formatting.

Implemented audio feature extraction with librosa and added waveform previews with a responsive UI.

Bootstrapped the FastAPI backend and audio pipeline. Planned the data model for uploads and per-session analysis.

Designed prompt patterns for structure and precision. Sketched a lightweight evaluation harness for schema compliance.

Covered tokenization, context windows, embeddings, vector search, grounding, and safety. Established baseline evaluation concepts.

Completed the core curriculum. Built multiple small apps (auth, REST, CRUD) and strengthened Python and SQL fundamentals.