Service
LLM and WhatsApp automation: agents, webhooks and document workflows
I connect language models to the channels your customers already use — mostly WhatsApp — and build the plumbing that keeps those conversations and documents from getting lost.
Who this is for
- Businesses answering customer queries, appointments or payments on WhatsApp by hand.
- Teams whose documents arrive in WhatsApp groups and need to end up in proper storage.
- Product teams adding an LLM feature that has to be reliable, not just impressive in a demo.
Problems I solve
Staff answer the same WhatsApp questions all day.
A conversational agent on the WhatsApp Business API that handles the routine flows (queries, appointments, payments) and hands the rest to a person. I built client-facing systems like this on LLaMA and OpenAI APIs at Shivohini TechAI.
Long conversations blow the token budget.
Keep context across turns deliberately — what to carry forward, what to summarise, what to drop — instead of resending the whole history.
Files shared in group chats get lost, and downtime means missed files.
An ingestion service that pulls media from group chats into Google Drive, plus a recovery job that backfills anything missed during downtime or network failures. Built and deployed at Zuneko Labs.
What you get
- WhatsApp Business API integration: webhook ingestion, message routing, template management.
- Conversational agents on LLaMA or OpenAI APIs, with explicit context and token management.
- Retrieval-augmented generation (LangChain, FAISS or Pinecone) over your own documents.
- Document ingestion from chats to cloud storage, with backfill for missed files.
- Async processing with Redis + Celery so replies don't block on the model.
Technology, and where it fits
- Meta WhatsApp Business API
- Webhooks, templates and message routing.
- LLaMA, OpenAI API
- Conversational agents and structured dialogue flows.
- LangChain, FAISS, Pinecone
- Retrieval over documents (RAG).
- FastAPI, Redis, Celery
- Webhook receivers and async AI responses.
- PostgreSQL / Supabase
- Conversation history and user state.
- Google Drive API
- Destination storage for ingested documents.
Evidence
- Agentic Honey-Pot — AI counter-scam platform
GUVI AI Impact Buildathon 2026: LLM personas with strategy switching and real-time scam-probability scoring.
- WhatsApp automation at Shivohini TechAI and Zuneko Labs
Client-facing agents for queries, appointments and payments; document ingestion with recovery.
Trade-offs worth knowing
LLMs are not deterministic
The same question can get a different answer. For anything with money or health involved, I use structured flows for the critical steps and keep the model for the open-ended parts, with a human hand-off.
Meta sets the rules
Business-initiated messages need approved templates, and free-form replies are limited to a window after the customer's last message. The design has to work inside those rules.
Hosted vs. self-hosted models
Hosted APIs are fastest to ship; self-hosting LLaMA gives more control over data and cost at volume, but you then run the GPUs.
Limitations
- I can't guarantee an agent never gives a wrong answer — I can limit where it's allowed to improvise and log what it does.
- WhatsApp access and pricing are controlled by Meta; approvals for your business account are outside my control.
Frequently asked questions
Can the agent take payments?
Appointment and payment flows are built as structured steps, not free-form chat — the model handles the open-ended parts and hands off to deterministic flows where money is involved.
What happens when the agent doesn't know the answer?
It says so and hands off to a human with the full conversation context. Agents are configured to improvise only within defined limits, with logging of everything they do.
Do I need Meta's business verification?
Yes — WhatsApp Business API access and pricing are controlled by Meta, and business-account approvals are outside any developer's control. I'll guide the setup, but the approval is yours.
Hosted or self-hosted LLM?
Hosted APIs are fastest to ship. Self-hosting LLaMA gives more control over data and cost at volume, but you then run the GPUs. The choice depends on your volume and data sensitivity.
How we'd work together
- A short call to understand the problem, your current system and your constraints.
- A written scope: what will be built, what won't, and how we'll know it works.
- Build in small milestones you can review, with the code in your repository.
- Handover: documentation, deployment notes and a walkthrough.
Open to full-time roles as well as freelance and consulting work.
Discuss a project
Tell me what you're building and where it's getting stuck. Freelance, consulting and full-time conversations are all welcome.