Developer Builds Email-to-Agent Memory Pipeline for Sales Deal Continuity

A developer built Waada, a deal-continuity system designed to convert messy account histories from sources like Slack exports and call transcripts into structured, queryable memory for AI agents. The core pipeline normalizes all inputs into a single schema, uses idempotent ingestion with stable source IDs, and scopes memory to individual accounts. To measure effectiveness, Waada compares three context sources: raw CRM data, a chronological summary baseline, and its own memory-aware retrieval system. The evaluation results are deliberately mixed, with the developer noting the architecture creates a different context path rather than guaranteeing superior accuracy. Key engineering takeaways include treating model output as an external dependency, separating retrieval from generation, and designing prompt budgets as a core architectural concern.