Case studies
Builds that survived
production.
Three anonymized case studies. Full before/after transformations. References available under NDA.
Inbound Email Triage + LLM Routing
The problem
A growing SaaS team was drowning in email. One person spent 45 minutes every morning manually reading, categorizing, and forwarding messages to the right Slack channel. Urgent customer issues got buried under newsletter noise. Misrouted emails meant missed escalations — and the whole thing depended on one human's judgment before their first coffee.
FIG · before
Then noorflows rebuilt it
FIG · after
The result
Fully automated classification pipeline. Gmail triggers pull new mail, an LLM classifier scores confidence across 5 categories, and conditional routing pushes each message to the correct Slack channel. Low-confidence scores trigger human escalation instead of silent misrouting. Idempotent on email ID — no duplicates, ever.
Google Sheets <> CRM Bidirectional Sync
The problem
Sales lived in Google Sheets. Marketing lived in HubSpot. Neither system knew about the other's changes. Every Monday, someone spent two hours copy-pasting contacts between them — introducing typos, creating duplicates, and silently overwriting edits. By Wednesday, the two systems had drifted apart again. Nobody trusted either dataset.
FIG · before
Then noorflows rebuilt it
FIG · after
The result
Bidirectional sync engine with email + phone deduplication that survives whitespace and case variations. A change-log audit trail records who edited what, when, in which system. Weekly reconciliation reports catch any edge-case drift. The two systems now stay in lockstep without human intervention.
Webhook > PDF > Email Delivery
The problem
A webhook triggered PDF generation and email delivery. Except when it didn't. The PDF API would timeout under load, the email step would silently swallow errors, and nobody knew a report was missing until a customer emailed asking 'where is my report?' three hours later. No retry logic. No alerting. No audit trail. The team spent 3+ hours a month firefighting phantom failures.
FIG · before
Then noorflows rebuilt it
FIG · after
The result
Rebuilt pipeline with exponential-backoff retry on the PDF generation step, verified delivery via SES, and a dead-letter queue for unrecoverable failures. Every execution is audit-logged. Telegram alerts fire immediately on DLQ entries — the team knows within seconds if something fails, not hours. Failure is a first-class event, not an exception.
These are anonymized examples. Your project gets the same production discipline.