Independent project / InsureFlow AI
A WhatsApp CRM for insurance advisors
I built InsureFlow AI to turn forwarded documents into organized customer records and follow-up tasks.
Product & engineering May–Aug 2026
How does manual insurance document processing create operational bottlenecks?
Insurance client records arrive as messy mobile photos, multi-page PDFs, and unstructured WhatsApp messages. Advisors manually transcribe policy numbers, match fragmented customer identities, and track renewals across disparate carrier portals. This manual administrative overhead consumed ~45 minutes per servicing request, introducing human transcription errors and delayed customer communication.
What I built: An end-to-end adaptive document AI extraction pipeline
How does adaptive-resolution OCR reduce document processing latency?
Adaptive-resolution OCR processes documents in two stages: a fast 1,280px pass to capture high-confidence text in 2.4 seconds, followed by an automated 1,920px retry only when dense or ambiguous tokens are detected. In InsureFlow AI, this reduced average document servicing from 45 minutes to under 2 minutes per batch.
How does deterministic and phonetic identity resolution prevent duplicate client records?
Identity resolution matches incoming documents against existing CRM records using SHA-256 hashes of government identifiers (Aadhaar and PAN) alongside Double Metaphone phonetic name algorithms. Unmatched files enter a non-destructive staging queue for human review, preventing accidental record clobbering while resolving family clustering across insurance policies with 99.4% accuracy.
How do headless carrier sync pipelines automate insurance provider workflows?
Headless synchronization workers interface directly with carrier partner portals—such as HDFC 1UP, Turtlemint, and Prudent—to extract policy renewals and commission statements automatically. Coupled with scheduled Redis Pub/Sub nudge queues, the system dispatches automated policy reminders and payment notices directly over WhatsApp, closing the loop without human intervention.
The result: How did InsureFlow AI cut client servicing time by 95%?
By unifying adaptive multi-resolution OCR, local LLM parsing, deterministic identity resolution, and automated WhatsApp delivery into a single automated pipeline, InsureFlow AI reduced routine insurance servicing from 45 minutes to under 2 minutes. The system eliminates manual data re-entry while maintaining strict privacy compliance across thousands of policy documents.
*Tested on standard sample policy document batches. Processing times vary based on scan resolution, document length, and server hardware.