Agentic AI Accelerator
Prove value with AI safely on your data — then scale. Choose RAG Starter, Agentic Workflows, or BI Co-pilot. Includes guardrails, evaluation and cost tracking.
Tracks (pick one or combine)
RAG Starter
Vector store over selected content; retrieval policy; citation and safety filters; offline eval set.
Agentic Workflows
Task-specific agents that can call ADF/Functions/REST (e.g., triage failed jobs, draft RCA, create Jira).
BI Co-pilot
Natural-language Q&A over your semantic model with strict scopes and audit logging.
What’s included
- Data readiness & policy: PII handling, retention, retrieval scope, redaction.
- Guardrails: prompt templates, tools allow-list, rate limits, content filters.
- Evaluation harness: offline set, quality gates (precision/latency/cost).
- Telemetry: usage, cost, and drift monitoring; rollback strategy.
Reference diagram placeholder — Data sources / Vector store / Policy & guardrails / Tools (ADF/Functions/APIs) / Observability & eval.
Time-to-value
- Week 1: Use-case selection, policy and data scope, baseline eval.
- Week 2–3: Build & integrate tools/agents, wire telemetry and cost tracking.
- Week 4: UAT with offline eval set, go/no-go and next-step plan.
Prerequisites
- Access to data sources and permissions to create an index/vector store.
- Target environment for agent tools (ADF/Functions/APIs) and service principals.
- For BI Co-pilot: governed semantic model (we can help establish one).
Deliverables
- Working pilot for the chosen track with policy, guardrails and telemetry.
- Evaluation results and risk assessment with remediation plan.
- Reference architecture and backlog to scale.
Ready to try AI safely on your own data?