The context
Non-technical founders in India — coaches, consultants, and brokers doing ₹10L–₹1Cr/month — run their sales on WhatsApp by hand. Rian Infotech built waba_automation, a WhatsApp-native automation platform on the official Meta WhatsApp Cloud API that turns the inbox into an always-on sales engine: a shared inbox, approved-template broadcasts, keyword auto-replies, and drip sequences, all topped with an AI bot that qualifies every lead and an AI voice agent that calls the hottest ones. The whole product is whitelabel-ready through a single branding config file.
The problem, precisely
Slow follow-up. Leads arrive on WhatsApp at all hours, but non-technical founders reply manually, so high-intent buyers go cold before anyone gets back to them.
No qualification. Every inbound message is handled by hand with no consistent way to score interest, so teams can't tell hot leads from tyre-kickers or know who to prioritise.
Manual nurture. Follow-ups, broadcasts, and multi-step nurture sequences slip when run by hand, and there's no safe, automated way to keep contacts engaged over time.
Rigid tooling. Founders serve different niches and languages (English, Hindi, Hinglish), but off-the-shelf tools can't be rebranded or re-tuned per client without a code rewrite.
What we built ✓ verified in code
Shared WhatsApp inbox on Meta Cloud API
A real-time shared inbox built directly on the official Meta WhatsApp Cloud API (Graph API) — conversation list, message threads, text and media sends, mark-as-read, and 24-hour session-window enforcement. Webhooks are HMAC-verified and deduplicated for reliable, idempotent message handling.
Broadcasts, auto-replies, and drip sequences
A Meta template browser with broadcast sends, a keyword auto-reply engine (exact / keyword / starts-with matching with priority ordering and rate limiting), and a drag-and-drop drip-sequence builder. A cron processor advances sequence steps safely using FOR UPDATE SKIP LOCKED to prevent double-sends.
AI qualification bot with BANT scoring
An AI bot that auto-detects English, Hindi, or Hinglish, replies via an LLM, and scores every lead on Budget, Authority, Need, and Timeframe (1–5 each). Leads are graded Cold, Warm, or Hot, auto-tagged, and hot leads trigger a handoff template plus a human alert.
AI voice calls to hot leads
When BANT qualifies a lead as hot, the platform places an AI voice call using xAI Grok realtime over a Python LiveKit agent, routed through a Twilio SIP trunk to the phone network. The call confirms identity, acknowledges the pain, and offers a demo — with per-call USD/INR cost tracking, a monthly budget guard, and business-hours limits.
Whitelabel-ready in one config file
A single typed branding config drives company name, app name, bot identity, logos, booking URL, and even AI prompt tuning (ICP, disqualification keywords, budget threshold). Rebranding the product for a new client or vertical is a config change, not a code change — proven by a real-estate edition cloned from the same base.
How it works
- 1
Capture. An inbound WhatsApp message hits the Meta webhook, which verifies the HMAC signature, upserts the contact and conversation, and runs keyword auto-reply checks before anything else.
- 2
Qualify. The AI bot engine detects the language, generates an LLM reply, sends it over WhatsApp, then scores the lead on BANT and auto-tags it as cold, warm, or hot.
- 3
Escalate. When a lead scores hot, the platform fires a handoff template to alert a human and schedules an AI voice call, which a cron dispatcher places via LiveKit and Twilio.
- 4
Nurture. Contacts enrolled in drip sequences are advanced by a one-minute cron job that sends the next template when due and logs each step, keeping follow-up consistent and automatic.
The outcome
The result is a production Next.js 15 and Supabase application, plus a Python voice runtime, that turns a manual WhatsApp inbox into an always-on, multilingual sales engine — capturing inbound leads, qualifying them on BANT, escalating the hottest ones to an AI voice call, and nurturing the rest through automated sequences. Built with HMAC-verified webhooks, concurrency-safe cron processing, and a vertical-clone architecture, it spins up a new niche by changing config and adding a feature module, as demonstrated by the real-estate build.