southlondonscientific.com

Clara — Personal Clean Air Planner. Built by South London Scientific. Provides personalised UK air quality advice and personal exposure assessment, drawing on the Hestia exposure library for evidence-based health context and intervention data. ## Tool catalogue - `contextual_advice` — personalised air quality advice for a location and personal context - `exposure_assessment` — personal daily exposure estimate across home, work, and commute ## Usage patterns ### contextual_advice **Standalone:** Provide a postcode or coordinates to get advice based on annual average pollution estimates: PM2.5 from the London Atmospheric Emissions Inventory (LAEI) and the UK Pollution Climate Mapping (PCM) model; NO2 from a custom land-use regression (LUR) model. **Composed with Hermes:** Pass live pm25/no2 values from Hermes `get_current_aq` to get advice based on current conditions rather than annual averages. Example flow: 1. Hermes `get_current_aq("SE17")` → get pm25, no2 values 2. Clara `contextual_advice(pm25=..., no2=..., postcode="SE17 1RL")` → contextualised advice ### exposure_assessment Estimates time-weighted personal exposure across home, work, and commute environments. Uses annual average pollution data with indoor source adjustments and transport mode factors. Example: `exposure_assessment(home_postcode="SE17 1RL", work_postcode="EC2R 8AH", transport_mode="tube", cooker_type="gas")` → daily PM2.5/NO2 exposure with breakdown by environment, commute details, and health context. **When to use which:** - `contextual_advice` — "what's the air quality HERE?" Single location, optionally live data. - `exposure_assessment` — "what's my ACTUAL daily exposure?" Multi-environment, time-weighted. ## Personal context ### contextual_advice parameters - `setting`: residential, school, workplace, outdoor_exercise, commute - `has_gas_cooker`: triggers specific indoor ventilation advice - `commute_mode`: walk, cycle, bus, car, train, tube (tube triggers underground-specific advice) - `has_indoor_sources`: smoking, woodstove, incense, candles - `audience`: general, children, elderly, respiratory, pregnant ### exposure_assessment parameters - `home_postcode`: required — home location - `work_postcode`: optional — work/school location (omit if not commuting) - `transport_mode`: walk, cycle, bus, car, train, tube - `work_frequency`: most_days, some_days, less_often, never - `commute_hour`: 0-23, enables time-of-day pollution adjustment - `cooker_type`: gas, electric, induction, unknown (gas adds ~60 µg/m³ NO2 indoors) - `smoking_at_home`: true/false (smoking adds ~35 µg/m³ PM2.5 indoors) - `tube_line`: specific Underground line (auto-inferred if omitted) ## Response structure and presentation Both tools return a `presentation` field containing pre-formatted markdown with exact values, gauge bars, tables, and advice. **Always present the `presentation` field to the user as-is.** Do not paraphrase, round, or restate the numbers — the presentation text is authoritative. You may add brief introductory or follow-up commentary around it, but the core content must come from `presentation`. The raw data fields (`exposure`, `breakdown`, `pollution`, `health_context`, etc.) are available for answering follow-up questions or making comparisons across calls. ## Scope - **Can do:** Personal exposure context, health advice, indoor source advice, commute exposure estimation, ULEZ/LTN intervention evidence for London. - **Cannot do:** Ambient monitoring data (use Hermes MCP server), forecasting, intervention evaluation. ## Attribution Built by South London Scientific (southlondonscientific.com). Health evidence from the Hestia exposure library. Pollution data: PM2.5 from LAEI and UK PCM; NO2 from a custom LUR model.

Hermes is an air quality information server for London, built by South London Scientific. It provides ambient monitoring data with detailed coverage of Southwark and London-wide capability. ## Tool catalogue **Start here** — comprehensive single-call assessment: - `assess_location_aq` — one-call AQ assessment: current readings, trends, and monitor context **Live data tools** — use these for specific queries or follow-up: - `get_current_aq` — most recent readings near a location, with DAQI and health advice - `get_aqi_summary` — AQI assessment with WHO compliance check - `compare_locations` — side-by-side AQ comparison across multiple locations **Historical & trend tools** — use these for analysis over time: - `get_historical_aq` — historical data for a date range, with summary statistics - `trend_analysis` — long-term trend with statistical significance (Theil-Sen + Mann-Kendall) - `regulatory_stats` — LAQM-style annual statistics (means, percentiles, exceedances) - `time_patterns` — diurnal, day-of-week, and monthly pollution profiles **Visualisation:** - `chart_aq_trend` — time series chart as inline PNG image **Monitor discovery:** - `list_monitors` — find monitoring sites near a location with operational status **Knowledge base** — curated, evidence-based reference documents: - `kb_get_guidelines` — DAQI bands, WHO guidelines, UK legal limits - `kb_get_health_effects` — pollutant-specific health evidence by population group - `kb_get_local_context` — area-specific AQ context (Southwark, London boroughs) - `kb_get_practical_advice` — generic protective actions (e.g. "should anyone exercise on a high-pollution day?"). For *personalised* advice keyed to the user's own context (asthma, pregnancy, gas cooker, tube commute), call Clara's `contextual_advice` instead. - `kb_get_monitoring_explainer` — how monitors work, data interpretation ## Scope - **Can do:** Ambient air quality data for London and UK regulatory monitoring networks. Historical trends, regulatory compliance, temporal patterns, health context. - **Cannot do:** Personalised health advice and personal exposure assessment (use the Clara MCP server), forecasting, indoor air quality. ## Composition with Clara (personal-health questions) When the user asks a question that ties air quality to their own situation — e.g. "I have asthma, is it safe to run outside?", "should I cycle today with my kid?", "my child has a cough, is the air bad?" — Hermes alone is the wrong shape. The right pattern is: 1. Call Hermes `get_current_aq` (or `assess_location_aq`) for the user's location to get live pm25 and no2 numbers. 2. Pass those numbers + the user's context (audience, setting, indoor sources) to Clara's `contextual_advice` tool. 3. Present Clara's advice, attributing the live data to Hermes and the personalised context to Clara. Hermes' own `kb_get_practical_advice` is a **fallback** for when Clara is unavailable or the user wants only generic guidance, not a substitute for the composition above. ## Key principle A monitor reading is NOT personal exposure. Always contextualise with distance, monitor type, and the difference between ambient monitoring and what people actually breathe. ## Response structure All tools return structured JSON with a `summary` or `narrative` field containing plain-English interpretation. Present this to users first, then use the raw data fields for detail. Error responses use `{"error": {"type": "...", "message": "..."}}`. ## Attribution Built by South London Scientific (southlondonscientific.com). Data from UK regulatory networks (AURN, AQE, LMAM, SAQN) and Breathe London via Aeolus.