Software Engineer GEN AI, Assistant Vice President

Citi is the world's most global bank, serving clients in more than 160 countries. Citi is investing heavily in scaled, enterprise-grade generative and agentic AI that operates securely and reliably across a highly regulated financial infrastructure. You'll join the USCC Architecture and AI Engineering team, which owns an enterprise multi-agent AI platform that autonomously supports the software delivery lifecycle, from product analysis through to reviewed, deployable code.  As a Generative AI Engineer (AVP), you will build and scale production components of this platform. This is a hands-on individual contributor role for an engineer with a strong foundation in computer science, software engineering, and applied AI, ready to ship real production systems rather than one-off prototypes.  Responsibilities   Build and maintain production AI agents covering product analysis, requirements capture, UI prototyping, backlog generation, and automated code implementation with pull request creation.  Implement multi-agent workflows using the Google Agent Development Kit (ADK), including sub-agent orchestration, human-in-the-loop approval gates, and agent-to-agent (A2A) messaging.  Integrate Gemini and Anthropic Claude into agent pipelines via enterprise gateways, applying advanced prompt engineering and structured tool/function calling.  Connect agents to enterprise systems, including Engineering & AI Product Documentation, JIRA, and GitHub, using MCP-style tool interfaces.  Build RAG pipelines using vector search, embeddings, and clustering to ground agent responses in accurate enterprise context.  Build evaluation harnesses (LLM-as-judge, parallel jury methods) to measure faithfulness, relevancy, hallucination rate, and task completion.  Write advanced, async Python and build well-documented FastAPI services, backed by persistent PostgreSQL session state.  Apply data structures, algorithms, and complexity analysis to keep orchestration and retrieval logic efficient at scale.  Instrument services for observability, maintain CI/CD and containerized deployments, apply secure coding and guardrail practices, and write tests (unit, integration, end-to-end) as part of every change.  Core Platform & Application Development   Full-Stack Engineering & Microservices: Lead the design, implementation, and lifecycle of custom AI platform components. Build production-grade, highly available microservices and full-stack applications using Python, Java (Spring Boot), and modern frontend frameworks (React) for internal LLM/SLM fine-tuning control planes, inference gateways, and experiment tracking dashboards. Agentic Frameworks & Patterns: Code, evaluate, and optimize multi-tenant intelligent agents using modular orchestration patterns such as ReAct (Reasoning + Acting) and ReWOO (Reasoning WithOut Observation), leveraging frameworks like Google's Agent Development Kit (ADK), LangGraph, and LangChain.   Strict Decoupled Integration Layers: Develop and enforce clean, decoupled enterprise integration architectures. Software Engineering Fundamentals & Distributed Systems  System Architecture & Design Patterns: Apply clean architecture, domain-driven design (DDD), and SOLID principles across Python and Java codebases. Implement robust design patterns (Strategy, Factory, Adapter, Circuit Breaker, Saga) for modularity and fault isolation.   Concurrency & High-Throughput Processing: Write robust concurrent and asynchronous code (e.g., Python asyncio, multithreading, Java CompletableFuture/virtual threads, connection pooling) to handle streaming tokens, parallel agent tool execution, and high-concurrency LLM inference loops.   Reliability, Fault Tolerance & Resiliency: Design systems for high availability, fault tolerance, and graceful degradation, including rate limiting, token bucket algorithms, retry policies with exponential backoff, and distributed caching strategies (e.g., Redis).   Enterprise API Design: Design clean, versioned, idempotent RESTful APIs, gRPC endpoints, and WebSockets (for real-time streaming LLM outputs) following OpenAPI / Swagger specifications.     Advanced Data Retrieval & Logic Engineering  Next-Generation RAG Architectures: Build and optimize end-to-end data ingestion and retrieval pipelines for advanced RAG paradigms, including Knowledge Graph RAG (GraphRAG), LightRAG, and hierarchical summary trees (RAPTOR).   Vector & Graph Integrations: Develop seamless, performant integrations with graph databases (Neo4j) and vector databases (pgvector, Pinecone, Milvus) for complex semantic, hybrid, and relational data retrieval.   Prompt & Intent Engineering: Design deterministic LLM system instructions, classification logic, and semantic routing trees to prevent intent collisions and hallucinations in complex multi-step workflows.   4 Technical Skills & Core Competencies   A. Software Engineering Fundamentals (Core Pillars) Data Structures & Algorithms: Strong command of core data structures (trees, graphs, hash maps, queues, heaps) and algorithmic complexity (Big-O analysis for time and space). Object-Oriented & Functional Programming: Good understanding of SOLID principles, clean code, design patterns, separation of concerns, and dependency injection. Distributed Systems Concepts: Understanding of CAP theorem, event-driven architectures, asynchronous messaging (Kafka / RabbitMQ), idempotency, distributed transactions, and database isolation levels. Testing & Quality Assurance: Comprehensive testing discipline across all layers: Unit testing (pytest, unittest, JUnit, Mockito Integration testing, contract testing, and mocking Test-Driven Development (TDD) principles LLM evaluation and regression testing frameworks CI/CD & DevOps Practices: Proficiency with Git workflows, automated CI/CD pipelines, containerization (Docker), container orchestration (Kubernetes / OpenShift), and infrastructure as code. Observability & Telemetry: Hands-on exper