Quick reference for each ADLC phase with key activities.
Phase 1: Prepare APIs and Data
Set up data sources, APIs, and MCP servers:
- Configure MCP servers (banking domain or custom)
- Set up vector database for RAG (if needed)
- Document and expose APIs
Phase 2: Build Agents
Implement your agent with framework, tools, and knowledge:
- Choose framework (Agno, LangChain, etc.)
- Code agent logic and instructions
- Connect tools (MCP servers or APIs)
- Set up RAG if needed
- Create FastAPI endpoints
Phase 3: Evaluate Agents
Test performance, safety, and quality before deployment:
- Run unit tests
- Evaluate with LLM judges (Langfuse)
- Human review for complex cases
- Red teaming (security tests)
- Measure metrics (latency, cost, accuracy)
Never skip evaluation. Agents must pass safety checks before production.
Phase 4: Run Agents
Deploy to production and expose via APIM:
- Deploy via CI/CD (auto on merge) or Helm
- Configure APIM endpoint
- Set up monitoring dashboards
- Configure alerts
Continuous: Monitor and Govern
Ongoing monitoring, security, and compliance:
Monitoring:
- Track performance (latency, cost, errors)
- Review traces for issues
- Set up alerts (PagerDuty, Grafana)
Governance:
- Enforce guardrails (AI Gateway, Nemo Guardrails)
- Run security scans regularly
- Periodic red teaming
- Compliance audits
Quick Reference
| Phase | Key Output | Next Phase |
|---|
| Prepare | MCP servers ready, data accessible | → Build |
| Build | Agent code complete, runs locally | → Evaluate |
| Evaluate | Tests pass, scores meet thresholds | → Run |
| Run | Deployed, APIM exposed, monitoring active | → Monitor & Govern |
Next Steps
