Portfolio-intelligence assistant for property energy management. Exposes tools to read live portfolio data (properties, annual energy/cost/CO₂ stats, energy-mix breakdown, forecasts, anomaly scans, and report generation) via HTTP endpoint or LLM-driven agent loop.
Techem Horizon has 13 tools with complete input schemas and descriptions, but quality is inconsistent. Naming follows verb_noun convention (list_, get_, rank_, etc.), which is good. However, descriptions are often terse (10-50 chars) and lack context on WHEN to use each tool or what downstream actions they enable. Parameter descriptions exist but are minimal. Output schemas are not documented, LLMs cannot predict response structure. Error handling is absent; tools return bare error dicts without recovery guidance. No tool annotations (readOnlyHint, etc.). The codebase shows proper parameter typing and filtering logic, but the tool interface lacks the LLM-optimization patterns needed for reliable agent composition.
Find properties with unusual annual kWh/unit (|z| > 1.0).
Generate the full portfolio intelligence report for the sidebar panel. Returns a handle the caller should pass through to the final response; the LLM receives a short confirmation, not the full report body.
Return currently open portfolio alerts (mold risk, heating failure, etc.).
12-month forecast for a representative unit (floor 1, apt A). HDD × linear.
Full detail for one property: metadata + annual stats.
Return the demo 'today' date used by the forecasting pipeline.
Output schemas not documented. LLMs cannot predict response structure (fields, types, nesting). Forces agents to guess or fail on unexpected formats.
Descriptions are terse (10-50 chars) and lack context. Missing WHEN to use each tool, what it returns, and how it chains with others. E.g., 'Return the demo today date' does not explain why an agent would call it.
No error handling or recovery guidance. Tools return bare error dicts (e.g., {'error': 'property X not found'}) without suggesting next steps or alternatives.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Return a filtered list of properties with unit counts and location.
Portfolio-wide aggregate: totals, averages, energy mix.
Rank properties by a metric. by ∈ {annual_energy_kwh, annual_cost_eur, annual_co2_kg}.
Return currently open portfolio alerts with priority levels and property/unit details.
Detect anomalies by scanning properties on annual kWh/unit and flagging high/low outliers (|z| > 1.0).
Generate the full portfolio intelligence report for the sidebar panel with sections on energy mix, efficiency, retrofit candidates, and forecasts.
Portfolio summary with inline answer blocks showing total energy, cost, emissions, and top consumers.
Duplicate tool definitions. Both llm_tools.py and tools.py define similar tools (e.g., anomaly_scan vs tool_detect_anomalies, generate_portfolio_report vs tool_generate_report). LLMs will waste reasoning cycles choosing between them.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All tools are read-only but this is not declared, limiting agent safety and optimization.