Personal project
WhatsApp Multi-Agent Ops
Turn a WhatsApp chat into a coworker who runs your SQL queries and hits your APIs for you.
- Result
- Ops questions answered in chat instead of a SQL editor.
- Stack
- Node.js · LangChain · Baileys · Gemini · DeepSeek · Vector Embeddings
- Links
- Source Code
API Agent Demo
Overview
Constantly switching between SQL editors and internal dashboards just to answer simple ops questions got old fast. So I built a bot you talk to like a coworker: ask a question in plain WhatsApp chat, and it routes the request to the right specialized agent to actually execute it.
Built early, when agentic AI tooling was still emerging, mostly as a hands-on way to learn the mechanics. Worth noting this predates standardized tool-calling protocols like MCP: the registries, routing, and context retrieval here are all hand-rolled, work a protocol and a platform like OpenClaw would now handle for me.
Highlights
- 01Multi-agent architecture using LangChain, Gemini, and DeepSeek for dynamic request routing and task delegation.
- 02Retrieval-Augmented Generation (RAG) with MiniLM embeddings to pull only relevant API/DB registries into context.
- 03Secure SQL execution restricted to predefined, human-reviewed queries with a 60-second confirmation window before anything runs.
- 04Self-healing validation-and-retry loops that catch and correct malformed structured outputs before execution.
- 05Runs entirely on free-tier models for zero-cost daily operation, with a simple swap-in path to paid models when needed.
Want something like this built?
Work with me