Project description VR-120391

Join our Development Center in Bucharest, and become a member of our open-minded, progressive and professional team. In this role you will be working on projects for one our world famous clients.

You will have a chance to grow your technical and soft skills, and build a thorough expertise of the industry of our client.

On top of attractive salary and benefits package, Luxoft will invest into your professional training, and allow you to grow your professional career.

Responsibilities

Role Summary:

Work in a scaled Agile working environment

Be part of a global and diverse team

Contribute to all stages of software development lifecycle

Participate in peer-reviews of solution designs and related code

Maintain high standards of software quality within the team by following good practices and habits

Use frameworks like Google Agent Development Kit (Google ADK) and LangGraph to build robust, controllable, and observable agentic architectures

Assist in the design of LLM-powered agents and multi-agent workflows (planning, tool use, orchestration, memory, and human-in-the-loop)

Lead the implementation, deployment and test of multi-agent systems

Mentor junior engineers on best practices for LLM engineering and agentic system development

Drive technical discussions and decisions related to AI architecture and framework adoption

Proactively identify and address technical debt and areas for improvement in AI systems

Represent the team in cross-functional technical discussions and stakeholder meetings

Key Responsibilities:

Design and build complex agentic systems with multiple interacting agents

Implement robust orchestration logic (state machines / graphs, retries, fallbacks, escalation to humans)

Implement RAG pipelines, tool calling, and sophisticated system prompts for optimal reliability, latency, and cost control

Apply core ML concepts to evaluate and improve agent performance, including dataset curation and bias/safety checks

Lead the development of agents using Google ADK and/or LangGraph, leveraging advanced features for orchestration, memory, evaluation, and observability

Integrate with supporting libraries and infrastructure (e.g., LangChain/LlamaIndex, vector databases, message queues, monitoring tools) with minimal supervision

Define success metrics, build evaluation suites for agents (automatic + human evaluation), and drive continuous improvement

Curate and maintain comprehensive prompt/test datasets; run regression tests for new model versions and prompt changes

Deploy and operate AI services in production, establishing CI/CD pipelines, observability, logging, and tracing

Debug complex failures end-to-end, identifying and document root causes across models, prompts, APIs, tools, and data

Work closely with product managers and stakeholders to shape requirements, translate them into agent capabilities, and manage expectations

Document comprehensive designs, decisions, and runbooks for complex systems

Skills

Must have

Education & experience

3+ years of experience as Software Engineer / ML Engineer / AI Engineer, with at least 1-2 years working directly with LLMs in real applications (not just experiments or coursework)

Bachelor's or Master's degree in Computer Science, Engineering, Mathematics, or a related field (or equivalent practical experience) Core technical skills Programming & software engineering:

Strong proficiency in Python (core language features, packaging, testing, async, type hints)

Very strong software engineering practices: version control (Git), unit/integration testing, code reviews, CI/CD

Experience building and consuming REST/gRPC APIs and integrating external tools/services Machine Learning (good understanding):

Understanding of core ML concepts: supervised/unsupervised learning, train/validation/test splits, overfitting, regularization, and common metrics (precision, recall, F1, ROC-AUC, etc.)

Good understanding of deep learning basics (neural networks, embeddings) and at least one ML/DL framework (e.g., PyTorch, TensorFlow, JAX, scikit-learn) LLMs & agentic AI (very strong understanding):

Deep practical knowledge of large language models:

Tokenization, context windows, temperature, top-p, system vs user prompts

Prompt engineering patterns (ReAct, chain-of-thought, tool-calling/tool-use)

Fine-tuning / adapters / instruction-tuning, or experience with RAG as an alternative

Experience building LLM-powered applications end-to-end: from idea → prototype → production

Familiarity with safety and reliability considerations: hallucinations, guardrails, content filtering, privacy Agentic frameworks (required understanding, experience preferred):

Conceptual understanding of modern agentic frameworks and patterns (stateful graphs, multi-agent coordination, human-in-the-loop, memory, and evaluation)

Hands-on experience with at least one of: o Google Agent Development Kit (ADK)

building multi-agent workflows, using its orchestration, tools, and evaluation features o LangGraph

designing graph-based, stateful agent workflows with cycles, branches, and durable execution

Candidates must be able to read, reason about, and extend ADK/LangGraph-based codebases

Direct production experience with both ADK and LangGraph is a strong plus Data & infra:

Experience working with vector databases (e.g., Pinecone, Weaviate, pgvector, Chroma) for retrieval-augmented generation

Comfortable with SQL and basic data modeling

Experience deploying on at least one major cloud platform (GCP, AWS, Azure) and using managed services (e.g., serverless runtimes, container orchestration, secrets management) Soft skills:

Ability to translate ambiguous business requirements into concrete technical designs

Strong communication skills; able to explain trade-offs to both technical and non-technical stakeholders

Comfort working in an experimental environment with rapid iteration, but with a strong bias towards production quality and maintainability

Nice to have

Experience with:

Vertex AI / Gemini or other hosted LLM ecosystems

Related frameworks and tools: LangChain, LlamaIndex, semantic search, evaluation frameworks (e.g., RAGAS, custom eval harnesses)

Monitoring and observability stacks (OpenTelemetry, Prometheus/Grafana/NewRelic, Datadog, etc.)

Background in one or more of:

Information retrieval / search

NLP (beyond LLMs): classic text processing, embeddings, semantic similarity

Security & compliance for AI systems (PII handling, access control, audit logging)

Contributions to open-source AI projects, blog posts, or talks about LLMs/agentic systems

Other

Languages

English: C2 Proficient

Seniority

Senior