Google ADK, Agentic AI, FDE
Skills Required: Python, Google ADK, Agentic AI, LangChain, LangGraph, Knowledge of Database and SQL
About the role
We’re delivering a large agentic AI program for a Tier-1 US bank: hundreds of production AI agents that automate compliance quality-assurance reviews across lending, consumer, cards, commercial and wealth lines of business. The agents are built on the bank’s agentic platform using Google’s Agent Development Kit (ADK) and Gemini on Vertex AI. Each one is validated in production until it meets a measured accuracy bar of 90% or better and is signed off by the bank’s Product Owners.
As Solution Architect, you’ll own the technical design of a pattern-based automation factory. You’ll turn the bank’s discovery outputs into reusable agent patterns, guide eight build pods through design, and make sure every agent is accurate, auditable and production-grade inside a regulated environment.
What you’ll do
- Design agents on Google ADK: decide where LLM agents apply judgment and where workflow agents and deterministic tools run fixed checks. Define state, tools, callbacks and structured outputs for each review type.
- Own the pattern library: define reusable patterns (system-of-record cluster + validation type), shared connectors and components, and decide for each automation whether it’s a pattern build or a pattern instance.
- Check technical readiness: review the bank’s discovery packs (procedures, process markdown, SOR lists, document types, historical records) and send gaps back before build starts.
- Lead system-of-record integration: design data access across origination, servicing, core banking and document systems, including cross-system keys, lineage, query performance and handling of missing sources.
- Design document intelligence: build extraction and classification for large loan and account files using Gemini and Document AI, with page-level citations and confidence-based routing to human review.
- Define and measure accuracy: co-author per-agent acceptance cards (attribute-level accuracy, critical-exception recall, sample sizes, exclusions), and build evaluation harnesses with ADK evaluation and Vertex AI Gen AI evaluation.
- Lead production tuning: run root-cause analysis on failed cases (data, rules, extraction, workflow), prioritize fixes, and make sure every change is regression-tested against gold sets.
- Build in banking controls: least-privilege IAM, PII handling, versioned prompts and models under change control, audit-ready run records, and evidence for model-risk management reviews.
- Engineer for scale and cost: design for full-population monthly runs, covering cost per case, caching, latency, resilience to SOR outages, and observability through Cloud Logging and Monitoring.
- Use AI-assisted engineering: set guardrails for AI coding agents (e.g., Devin) and Gemini Code Assist, including pull-request-only changes, sandboxed data and human review.
- Lead and collaborate: run design reviews with onsite Tech Leads across pod pairs, mentor developers, and work directly with bank Product Owners, platform teams, SOR owners and model-risk stakeholders.
What you’ll bring (must-have)
- Agentic AI in production: hands-on experience designing and shipping LLM-based agents, not just prototypes, including evaluation, monitoring and iteration after go-live.
- Google ADK and Vertex AI: practical depth with ADK (LLM agents, workflow agents, tools, session state, callbacks, evaluation) and Gemini models on Vertex AI.
- Google Cloud architecture: Cloud Run, BigQuery, Cloud Storage, Pub/Sub, Secret Manager, IAM, VPC Service Controls, and Cloud Logging and Monitoring.
- Engineering: strong Python and advanced SQL against enterprise databases; API integration; structured outputs and schema validation; CI/CD and Git-based workflows.
- Evaluation discipline: gold datasets, attribute-level metrics, precision and recall trade-offs, LLM-as-judge calibration, and drift detection.
- Regulated industry experience: delivery in banking or financial services, ideally in compliance, quality assurance, lending operations, KYC/AML or audit, with working knowledge of model-risk governance (e.g., SR 11-7) and data-privacy controls.
- Design leadership: proven ability to set reusable architecture standards across multiple delivery teams and to defend design decisions with business stakeholders.
Nice to have
- Vertex AI Agent Engine for deploying and managing ADK agents; Document AI processors.
- Agent interoperability standards (e.g., MCP, A2A) and multi-agent orchestration.
- Mortgage or lending compliance knowledge (e.g., TRID disclosures and fee tolerances), or cards and payments disputes.
- Process-intelligence or task-mining tools for discovery.
- Certifications: Google Cloud Professional Cloud Architect, Professional Machine Learning Engineer.
.jpg?format=webp)
