Overview
I led the end-to-end delivery of DAISI (Decathlon AI Supplier Informations), a production Enterprise Agent in Google Chat for supplier and finance workflows. The work began with client discovery meetings and process mapping, then moved through architecture, Workday Finance and business-system integrations, implementation, evaluation, security validation, production rollout, and adoption. DAISI combines agentic business tools, governed RAG, Gemini, Vertex AI Vector Search, Cloud Run, Cloud SQL, MLflow/Databricks, Model Armor, DLP, OpenTelemetry, and Terraform. It was designed for a 10,000-user target population and successfully load tested with 2,000 concurrent users. The application runs in production with automated infrastructure management and end-to-end observability. Its reported annualized impact is 13,000 hours/year saved for supplier-support workflows.Business Impact
| Metric | Value |
|---|---|
| Target population | Designed to support 10,000 users |
| Reported annualized impact | 13,000 hours/year saved for supplier-support workflows |
| Load test | Successful test with 2,000 concurrent users |
| Availability | 24/7 instant answers vs. waiting for human response |
| Coverage | Grounded on governed supplier and finance procedures |
| Escalation | Automatic routing to the appropriate human expert when needed |
My Ownership
- Led client discovery meetings, requirements clarification, architecture decisions, and delivery planning
- Built the ReAct-style LangGraph agent, governed retrieval, tool interfaces, persistence, security controls, and cloud runtime
- Integrated Workday Finance and other approved business services behind controlled tool boundaries
- Designed A2A-ready interoperability interfaces for reusable cross-agent capabilities
- Built an internal Agent Factory with shared libraries, templates, evaluation, observability, security, and deployment conventions, improving delivery time-to-market by up to 5×
- Designed integration tests and detailed readiness reports covering functional behavior, reliability, security, latency, cost, and scale
- Optimized caching, model usage, asynchronous processing, autoscaling, and observability for production cost and latency
- Presented production agent-development practices to 100+ colleagues and contributed to technical enablement through training, documentation, onboarding material, and runbooks
Technical Architecture
AI/ML Stack
- LangGraph - Agent orchestration framework with stateful conversation flows
- LangChain - LLM application framework for chains and prompts
- Gemini via Vertex AI - Foundation model for text generation
- Vertex AI Vector Search - High-performance semantic search for RAG
- FAISS - Local vector similarity search
- Model Armor - GCP security templates for anti-prompt injection and content filtering
- DLP - Protection for sensitive information
Backend Infrastructure
- FastAPI - High-performance async API framework
- Cloud Run - Serverless container deployment with autoscaling
- Cloud SQL PostgreSQL - Managed database for conversation persistence
- Google Cloud Storage - Object storage for knowledge base and configurations
- Uvicorn - ASGI server for FastAPI
Observability & MLOps
- MLflow 3.7+ - Experiment tracking with LangGraph autologging
- Databricks - MLflow and Delta refresh workflows
- LiteLLM - Unified LLM API gateway
- OpenTelemetry - Distributed tracing with OTLP/gRPC export
- Cloud Logging - Centralized log aggregation
Operational Automation
Automated Cloud Run jobs and scheduled workflows handle lifecycle operations such as retention enforcement, trace evaluation, and governed knowledge synchronization. The public case study intentionally omits internal schedules, identifiers, and source-system details.Infrastructure as Code
- Terraform - Modular management of the runtime, scheduled workloads, identity boundaries, security controls, retrieval infrastructure, and storage
DevOps & CI/CD
- Docker - Containerized deployments
- GitHub Actions - Automated CI/CD pipelines
- SonarCloud - Code quality and security analysis
- Pre-commit hooks - Automated code checks
- Ruff - Python linting and formatting
- Mypy - Static type checking
- Pytest - Async test framework with coverage
Project Management
- Jira - Sprint planning and issue tracking
- Confluence - Technical documentation with auto-sync from /docs
Public Architecture View
The public architecture is intentionally sanitized. At a high level, Google Chat events enter a FastAPI runtime on Cloud Run, pass through authentication and input-security controls, and reach a LangGraph agent. The agent can retrieve governed evidence, call approved business tools, preserve conversation state, and expose A2A-ready interface boundaries. Output controls are applied before the response returns to the user, while traces and evaluation signals are recorded asynchronously. This separation keeps the system understandable and testable:- Channel and API boundary for authentication, idempotency, and response delivery
- Security boundary for prompt-injection protection, sensitive-data controls, and policy enforcement
- Agent runtime for planning, retrieval, tools, memory, and A2A-ready interoperability
- Enterprise integration boundary for approved finance and supplier services
- Observability boundary for traces, quality evaluation, latency, cost, and operational reporting
Core Features
Intelligent Question Answering
RAG-powered responses grounded in operational and accounting practical guides ("fiches pratiques").Contextual Disambiguation
Multi-turn conversations with follow-up questions to clarify ambiguous requests.Conversation Memory
PostgreSQL-backed checkpointer for persistent conversation state with IAM authentication.System Integrations
- Identity context - Authorized user context used to scope responses and tools
- Workday Finance - Controlled integration for approved finance workflows
- Human-expert routing - Escalation to the appropriate support contact
- Supplier and purchasing tools - Approved APIs exposed through typed agent tools
- A2A-ready interfaces - Designed boundaries for reusable governed capabilities
Security & Compliance
- Model Armor templates for prompt injection protection
- DLP to protect sensitive information
- Policy-based handling of prohibited and out-of-scope topics
- GDPR-compliant data retention with automated TTL cleanup
- Identity-based access control and least-privilege service boundaries
- Integration and security test reports used as production-readiness evidence
Technologies Summary
| Category | Technologies |
|---|---|
| AI/ML | LangGraph, LangChain, Gemini, Vertex AI Vector Search, FAISS, Model Armor, DLP |
| Backend | Python 3.11, FastAPI, Uvicorn, Pydantic |
| Database | Cloud SQL PostgreSQL, langgraph-checkpoint-postgres, psycopg3 |
| Cloud | GCP, Cloud Run, Cloud Run Jobs, Cloud Scheduler, GCS |
| Observability | MLflow, Databricks, LiteLLM, OpenTelemetry |
| Infrastructure | Terraform, Docker |
| CI/CD | GitHub Actions, SonarCloud, Pre-commit |
| Quality | Ruff, Mypy, Pytest, pytest-asyncio |