Operational Metrics Claude Agent
Budget: $15 – $25 USD
I need an Anthropic Claude–powered agent that ingests our operational data and turns it into clear, actionable performance-efficiency insights. The agent’s sole purpose is data analysis: no chit-chat, just hard numbers explained in plain language.
Scope
• Connect to the raw operational metrics we export daily (CSV or simple API feed—your choice, just document it).
• Parse, clean, and structure the data so Claude can reason over it reliably.
• Prompt-engineer the model to surface trends, efficiency scores, and any outliers that signal a drop in performance.
• Return a concise report (JSON + human-readable summary) that highlights:
– Week-over-week efficiency movement
– Key drivers behind any spikes or dips
– Suggested actions when performance flags fall below preset thresholds
Deliverables
1. A runnable script or lightweight app (Python preferred) that calls Claude via API, performs the analysis, and outputs the reports.
2. Configuration file for data source paths, API keys, and threshold settings.
3. README with setup steps, sample command, and one worked example using dummy data.
4. Two short Loom videos: one that shows installation, another that walks through a live analysis.
Acceptance Criteria
• Feeds a 50 MB CSV in under 30 seconds.
• Identifies at least 90 % of synthetic anomalies in a provided test set.
• Summary section stays under 250 words yet references every major variance detected.
If you’ve shipped similar analysis bots or have strong prompt-engineering chops with Claude, let’s get this rolling.
Scope
• Connect to the raw operational metrics we export daily (CSV or simple API feed—your choice, just document it).
• Parse, clean, and structure the data so Claude can reason over it reliably.
• Prompt-engineer the model to surface trends, efficiency scores, and any outliers that signal a drop in performance.
• Return a concise report (JSON + human-readable summary) that highlights:
– Week-over-week efficiency movement
– Key drivers behind any spikes or dips
– Suggested actions when performance flags fall below preset thresholds
Deliverables
1. A runnable script or lightweight app (Python preferred) that calls Claude via API, performs the analysis, and outputs the reports.
2. Configuration file for data source paths, API keys, and threshold settings.
3. README with setup steps, sample command, and one worked example using dummy data.
4. Two short Loom videos: one that shows installation, another that walks through a live analysis.
Acceptance Criteria
• Feeds a 50 MB CSV in under 30 seconds.
• Identifies at least 90 % of synthetic anomalies in a provided test set.
• Summary section stays under 250 words yet references every major variance detected.
If you’ve shipped similar analysis bots or have strong prompt-engineering chops with Claude, let’s get this rolling.