Software Engineer 2 (Python) - maternity cover

In our domain, we focus on delivering the core search and retrieval capabilities for the Allegro AI Assistant. We maintain a critical entry point for users looking to discover products, compare offers, and get personalized recommendations through natural conversations. Striving for relevancy and accuracy first, we are designing, implementing, and operating an extensive system combining modern microservice architecture with Large Language Models, high-performance retrievers, and vast data sources. We care about data quality, robust context building, and near real-time accuracy to best guide users through a highly dynamic marketplace via an intelligent, conversational interface.

• Code in Python, Java and Kotlin.
• Expect work that requires creative approach to search experience challenges.
• Want to create solid code, accept comments, and share knowledge during code reviews.
• Can look for effective solutions to user requirements.
• Want to constantly develop and update knowledge.
• Know English at at least B2 level.

Nice to have:

• Basic experience with LLM tuning (LoRA, PEFT), training and evaluation pipelines for models.
• Experience with frameworks for building AI Agents (Agent Development Kit, LangChain, AutoGen, CrewAI).

Performance & Scale:

• Design, build, and optimize AI-driven systems that process complex natural language interactions.
• Deliver low-latency, high-quality responses from our AI Assistant requiring high-availability and smart optimization of LLM and retrieval workloads.

Modern Tech Stack:

• Work with a cloud-native polyglot environment including Java, Kotlin, Spring, ADK, and Python.
• Utilize Python and LangChain for complex LLM orchestration, alongside Langfuse for observability, prompt tracking, and evaluation.
• Leverage advanced vector search and modern ML solutions.

True Ownership & Autonomy:

• Join an agile, autonomous team with full ownership of AI microservices and data pipelines from design to deployment.
• Improvements dictated by results of experiments and AI evaluation metrics.

Complex Challenges:

• Work on LLM orchestration, Retrieval-Augmented Generation (RAG), and semantic search optimization.
• Integrate advanced embedding models, prompt pipelines, and vector databases across Java/Kotlin and Python ecosystems.