Claude Operational Data Analyst

The goal is to build a Claude-powered agent that can ingest our day-to-day operational data, turn it into clear periodic reports, and surface forward-looking insights through predictive analysis.

Data scope
• Data sources include production logs, supply-chain metrics, and KPI spreadsheets stored in Google BigQuery and Google Sheets.
• Volumes are moderate—about 2 GB daily—so latency is less important than accuracy and clarity of the output.

Key capabilities I need baked into the agent
1. Scheduled and on-demand report generation in PDF and Markdown, summarising current operational performance against set KPIs.
2. Predictive analysis that flags potential bottlenecks or capacity issues at least two weeks in advance, using past trends.
3. A simple chat interface (web or Slack) so team members can ask follow-up questions and request custom slices of the data.

Tech preferences
Claude (latest model via Anthropic API) as the core LLM, Python for the orchestration layer, and a lightweight vector store (e.g., Pinecone or Chroma) for context retrieval. I’m open to alternatives if you can justify gains in speed, cost, or accuracy.

Deliverables
• Clean, well-commented code repository
• Deployment scripts (Docker or serverless)
• A short read-me explaining how to add new data sources and adjust reporting schedules
• One live demo session to walk through setup, outputs, and maintenance steps

Acceptance criteria
The agent must produce weekly summary reports without manual intervention and correctly predict at least 80 % of historically observed capacity crunches in a back-test over the last six months.

If everything above sounds clear, tell me how you would architect the pipeline and how long you’d need for a first working version.