Senior Gen AI / Agentic AI Engineer

The engineer, within the framework of the generated ideas for improving processes and implementing new features by the domain, must conduct RnD (PoC, MVP phases) to determine approaches to implementing use cases, provide architecture design and implementation based on research into the list of solutions for implementing AI use cases.
That will include:

• Design and development of multi-agent systems, orchestrators based on ReAct logic and Agentic RAG systems and reuse and refinement of existing approaches and architectures.
• Setting up the interaction of LLM, Function Calling (vendor and open source models) with internal banking services to perform real client operations.
• During the implementation process, if necessary, implement rigid business schemes and Human in the loop mechanisms to prevent model hallucinations, etc.
• Setting up systems for protecting against Prompt Injection, automatic masking of personal data and output filtering of responses (Fact-checking).
• If necessary, design long-term memory to preserve the context of previous conversations and user operations.
• Ensure the implementation of best practices to minimize hallucinations and ensure cost-effectiveness of models.

• Hands-on experience in developing and evaluating GenAI services, including RAG systems understanding “reAct” (Reasoning + Acting)+ Agentic RAG logic, integrating LLM APIs and prompt/context engineering, as well as training, fine-tuning and deploying machine learning models in production environments.
• Knowledge of machine learning and generative AI frameworks LangGraph or LangChain, PyTorch / TensorFlow, Hugging Face, OpenAI/Anthropic SDK.
• Hands-on commercial experience working with AI models via API (Gemini, Anthropic) and Self-hosted models (Llama 3, Mistral, Mixtral) and understanding and differences of models based on FR/NFR.
• Deep understanding of how LLMs interact with external APIs via Function Calling.
• Working with vector databases, knowledge of semantic search methods and principles of database structuring and cleaning.
• Practical experience with at least one cloud platform (AWS, GCP or Azure).
• Knowledge of Python and understanding of asynchronous programming.
• Understanding of the AI ​​model lifecycle: monitoring, versioning and quality assessment (RAGAS, DeepEval) and experience with these tools.
• Experience with Guardrails to set strict limits on the topics of conversations and agent actions, filtrations that automatically remove personal data before sending a request to external LLMs, as well as the ability to create Output filters that check the generated response for Fact-checking before displaying.
• Deterministic Logic Integration – launching the AI ​​agent to work according to strict schemes to prevent the model from “making up” it.
• Knowledge of approaches to automated testing of responses on large datasets to measure the percentage of hallucinations before releasing to MVP will be an advantage.
• Knowledge of approaches to implementing Human-in-the-loop mechanisms so that the agent cannot perform an action without final verification by the user.
• Ability to design memory systems that store the context of previous conversations and customer operations for personalization Long-term Memory & User Context.
• Experience in the Fintech domain would be a great advantage
• Ability to effectively communicate complex technical concepts and mentor team members.
• Ability to quickly test new libraries and approaches.
• Analytical Problem Solving, namely the ability to debug complex “black boxes” and understand why the agent behaves unpredictably.
• Focus on creating a working prototype, not a perfect scientific paper.