Salesforce Reporting AI Agent

Our administrative platform needs an AI agent that can tap directly into Salesforce, merge that data with our internal APIs and Postgres store, then surface clear, reliable reports for our operations team. The core objective is better data processing and richer reporting, not just task automation, so the agent must reason over live records, invoke LLM tools through function-calling or RAG when beneficial, and present insights in a format the team can approve or override before any sensitive change is committed.

You will work in Python (FastAPI powers most of our backend) and have full access to a staging environment where you can wire the agent into existing authentication, RBAC, and audit-log middleware. Every action the agent attempts—record creation, status update, file upload—must respect our current permission matrix and write a traceable log entry. Where a high-risk update is detected, the agent should automatically route a human approval request to our Slack workflow before proceeding.

Deliverables
• A containerised AI agent service that connects to Salesforce via REST/bulk APIs and our internal API gateway
• RAG or equivalent retrieval layer for long–form policy docs (stored in S3) to ground the agent’s responses
• End-to-end unit/integration tests plus a short demo notebook showing typical report generation and approval flow
• Deployment guide for our Kubernetes cluster along with environment variables and secrets layout

I’d like to see two or three similar agent projects you shipped to production, your preferred libraries or frameworks (LangChain, LlamaIndex, custom, etc.), and a realistic timeline. I’m ready to start as soon as we align on scope and milestones.