Three things we built for ourselves first.

01 · Problem

Our businesses run on WhatsApp. Staff turnover, phone resets and disappearing messages meant promises and details vanished — and we still had to answer regulator and customer requests for old records.

02 · What we built

We built an append-only, SHA-256-chained archive that mirrors the WhatsApp Business history into a vault, auto-links every contact to the CRM, and extracts GSTINs, PANs and emails from message bodies.

03 · Numbers

334,246 messages archived. 85% auto-linked to clients, 65% to leads. 13,866 entities extracted (617 GSTIN, 1,477 PAN).

04 · What broke

An @lid privacy-identifier mismatch silently dropped ~1,131 chats from the auto-link. Took two weeks to spot. Fix: added @lid → phone-number resolution + a chain-integrity verification script that runs in the daily anomaly cron.

05 · Today vs coming

Today: sync-from-phone, read-only, 334,246 messages live. In development: real-time webhook capture (Coexistence-mode limited on groups), an audited DPDP self-serve erasure tool, bulk voice-note transcription.

01 · Problem

Two businesses, two CRMs and two WordPress sites were stuck on shared cPanel hosting with no real disaster-recovery story — a single account closure could have wiped years of data.

02 · What we built

We lifted everything onto a self-managed cloud VPS behind Cloudflare in a single planned cutover, kept old URLs alive with 301s, and stood up encrypted backups with tested restores and cost monitoring.

03 · Numbers

Single planned cutover with 301 redirects. 8/8 checks on the restore drill. Billing-outage detection cut from 7 days to ~24 hours.

04 · What broke

Let's Encrypt cert install slipped to the day after cutover — Cloudflare's edge cert held the public side, but we ran a ~24h gap on direct origin TLS. We don't call the migration zero-downtime; we call it a single planned cutover.

05 · Today vs coming

Today: 2-of-5 backup layers active, 8/8 restore-drill verified, the billing canary rule live. In development: offsite replica + cold-storage archive + immutable air-gap to bring 2-of-5 up to 5-of-5.

01 · Problem

Knowledge lived in thousands of unsearchable chats, and good follow-ups silently slipped — a quote never chased, a renewal missed.

02 · What we built

We extract every conversation into a knowledge graph and surface 'next-best-action' cards on staff desks. Staff edit and send; the AI never messages a customer on its own.

03 · Numbers

Lead Desk in production. HTTP-503 failure rate dropped from ~50% to ~0 after moving extraction to Vertex AI. Knowledge writes ~23× faster.

04 · What broke

The autonomous-reply AI Worker was scaffolded and ~64KB of design committed before we discovered Coexistence-mode doesn't expose group chats to the WhatsApp API. We paused the entire autonomous path rather than ship half. It's on the roadmap, not the in-development list.

05 · Today vs coming

Today: Lead Desk in production, ~20+ anomaly rules live, Customer Intelligence Capsule v1 across surfaces. In development: Customer Care + Channel Partners desks. Roadmap: autonomous reply (gated on Meta WABA + group-API exposure).