Location: 100% Remote (German Normal Employment Contract, International EOR, or Contractor) Employment Type: Full-time Working Hours: Flexible Working & Company Language: English Start Date: Immediately or as agreed

About Us

We are a small, but growing SaaS technology startup building production-grade language understanding systems for hotels and travel companies. Our systems process real-world (inbound) emails and documents at scale and extract structured, actionable data from unstructured text.

This is not a chatbot and not a prompt-only project. Our platform performs single-pass inference, combining deterministic rules, configuration-driven logic, and ML/NLP models to reliably classify intent and extract information under real production constraints.

Why this problem is interesting and hard

Hotel emails are deceptively complex. For example, a single booking-intent email may contain:

• Arrival and departure dates, or a length of stay
• Number of guests
• Room types that imply single or double occupancy
• Children and their ages, which must be evaluated against hotel-specific child policies to determine whether a child counts as an adult

One email may even contain multiple booking requests, sometimes across different hotels, which must be parsed and interpreted independently.

To further complicate matters, interpretation depends on hotel-specific configuration, such as:

• When a child is considered an adult
• When a request qualifies as a group booking (e.g., 7 rooms) versus an individual reservation

This creates real engineering trade-offs between deterministic logic, ML-based inference, and LLM-supported components — all while meeting strict requirements for accuracy, throughput, cost, and data protection.

Where we are today

We run a live production system based primarily on custom-trained RASA models. Our current focus is on improving overall Precision, Recall, and F1, and evaluating whether newer model architectures can partially or fully replace existing components — without sacrificing speed, reliability, or commercial viability.

All systems run in AWS infrastructure, under strict GDPR and performance requirements.

Even though we’re technically deep, we’re also a friendly, collaborative startup. We have a strong error culture: blame is boring and useless. What matters is understanding why something happened and how, as a team, we can improve our systems so that the next mistake is a new one — not the same one again.

Our working and company language is English, and we collaborate daily in a fully international, remote-first setup.

Your Role

As an Applied NLP / Language Systems Engineer, you will work on the core language-processing pipelines that power our product.

You will:

• Design, implement, and improve intent recognition and structured information extraction pipelines
• Build single-pass inference systems that combine deterministic logic, configuration data, and ML models
• Select, fine-tune, evaluate, and deploy ML/NLP models in production
• Measure, and improve quality metrics such as precision, recall, F1, and error rates
• Build evaluation pipelines and monitoring to ensure stable production performance
• Collaborate closely with backend engineers and product stakeholders to deliver reliable, measurable results

This is a hands-on role with direct influence on architecture, modeling decisions, and evaluation methodology.

What We Are Not Looking For

• Prompt engineers
• Chatbot builders
• "API-only” AI integrations without systems understanding

We are looking for engineers who understand that ML and LLMs are components in a larger system, not magic black boxes.

Requirements

Must-Have

• Strong experience in Applied NLP / Machine Learning, with production system exposure
• Ability to design hybrid pipelines combining deterministic rules and ML models
• Solid understanding of evaluation metrics (precision, recall, F1) and error analysis
• Experience fine-tuning, deploying, or operating ML models in production environments
• Strong Python skills and experience with ML frameworks (e.g., PyTorch, TensorFlow)
• Awareness of privacy, compliance, and operational constraints (e.g., GDPR)

Nice-to-Have

• Experience with open-source LLMs or encoder-only models in production
• Familiarity with RASA, spaCy, or classical NLP frameworks
• Experience with retrieval-augmented systems or vector-based approaches
• Cloud or infrastructure experience (AWS, GCP, Azure)

What You’ll Get

• 100% Remote work — work from anywhere
• Flexible hours — truly pick your own hours - outside of a meeting or two a week
• Direct impact on core ML/NLP systems and architectural decisions
• A small, highly technical, pragmatic team with flat hierarchies
• Modern technical environment and provided equipment (laptop, monitor if needed)
• A strong error culture: blame is boring and useless. We focus on learning, improving systems, and making sure we make new mistakes — not repeat old ones
• Fair / Impact based Compensation - honestly, we're not yet at the point where we can shower you with money, thoug we're not cheaping out either. But if you can drive us above 90% F1, then you'll have earned a massive raise, because your improvements will have helped generate the revenue to pay for it.
• Base salary: competitive for a senior Applied NLP/ML engineer across Europe (approx. €50k–85k/year), depending on experience and contract format. Performance-linked: if you help push our F1 scores above 90%, you’ll earn a significant raise (up to 50%), because your improvements generate the revenue to make it happen.

Application

Please send:

• Your CV / resume

• Relevant project examples or technical descriptions

• A short note explaining why you are interested in working on this particular problem. What is it that make that challenge interesting?

  Email applications to: career@hotelresbot.com

For questions about the role or application process, feel free to reach out via email.

Bereitgestellt in Kooperation mit der Bundesagentur für Arbeit.