| Sumario: | Farmer call centers capture high-frequency, location-specific information needs and constraints, offering a farmer driven dataset for designing scalable agroclimatic advisory services. This progress report documents the development of an AI-enabled pipeline that converts India’s call center logs into operational agroclimatic intelligence for targeting, prioritization, and message refinement. Using on-disk processing (DuckDB) for multi-million-record repositories, the workflow standardizes dates and administrative locations, de-identifies sensitive fields, and applies multilingual natural language processing (NLP) to (i) classify calls into WEATHER_DIRECT, WEATHER_TRIGGERED_AG, and OTHER intents and (ii) map climate-related calls into interpretable topic groups (e.g., rainfall/monsoon, sowing/onset timing, irrigation and water management). Spatiotemporal aggregation produces routine monitoring indicators (annual and monthly demand trends, state year heatmaps, and district-level intensity distributions) and supports hotspot ranking and anomaly flagging for expert review. Service-quality metrics are prototyped (answer rate, response length, and a preliminary actionability proxy) to identify topics that benefit from standardized scripts, decision trees, and escalation protocols. Initial descriptive outputs demonstrate strong monsoon-driven seasonality, concentration of demand in a subset of states and districts, and topic dominance by rainfall/monsoon queries. These results are preliminary and will be refined through full-scale reprocessing, improved
text normalization, and validation with domain experts. The report concludes with an operational scale-up pathway human-in-the-loop playbooks and message templates integrated with Kisan Call Centre (KCC), IMD–GKMS advisories, and mKisan and an evaluation plan to estimate productivity and resilience impacts as the system moves from prototype to deployment.
|
|---|