Join a small, fully remote team building AI-powered backend systems for the Indian construction industry, working with WhatsApp bridges, LLM agents, and real-time messaging infrastructure.

About the Company

Pyrock AI (The Pyrockai LLP) runs construction sites through WhatsApp. Their AI assistant, Mahi, gets added to a builder's existing site WhatsApp groups and takes over the paperwork that nobody wants to do: auto-entry from PDF bills, voice notes in Hindi, Marathi, Gujarati, and English, live inventory from bills plus consumption, and a 3-line daily safety summary from site cameras. The owner asks a question in the group and gets an answer back — no app, no dashboard, no training.

They are live with paying builders in Mumbai and Ahmedabad and onboarding more sites every month. The engineering team is small, which means your code goes to production quickly and you see its effect on real sites within days.

Role Overview

We are looking for a Senior Backend Developer (AI Systems) to join our fully remote team. You will be responsible for building and maintaining the core backend infrastructure that powers our AI assistant — from the WhatsApp bridge that keeps our agent inside construction site groups, to the pipeline that processes messy multilingual input and returns reliable operational data.

This is a role for someone who enjoys owning complex systems end-to-end and thrives in a small, high-agency team. You will work directly with LLM agents, real-time messaging, and computer vision — a combination that is difficult to get exposure to at larger companies.

Key Responsibilities

• Build and maintain our WhatsApp Web bridge — the service that keeps a dedicated number logged in and live inside client site groups: session persistence, reconnection and QR re-auth handling, group event handling, media download, and message ordering under load. This is the layer everything else depends on, and keeping it stable is the single most important job on the team.
• Build and maintain the Pyrock agent pipeline — message ingestion, routing, context handling, tool calling, and reply generation.
• Work with Meta's WhatsApp Cloud API alongside the bridge: webhooks, templates, media handling, number provisioning.
• Build ingestion for messy real-world input — scanned PDF bills, WhatsApp voice notes across four languages, site photos.
• Write the guardrails that keep the system trustworthy: no hallucinated numbers, no dropped messages, graceful behavior under load.
• Deploy and monitor per-client infrastructure (isolated VPS per client), with proper alerting.
• Contribute to our computer-vision module — PPE and helmet compliance, gate material in/out, idle detection from existing site IP cameras.

Tech Stack

• Languages: TypeScript, Node.js
• Backend: Node.js, Express (implied)
• Messaging: WhatsApp Web bridge (whatsapp-web.js, Baileys, Venom), Meta WhatsApp Cloud API
• AI/LLM: LLM APIs (GPT, etc.), agent patterns (tool calling, structured outputs)
• DevOps: Linux, VPS, Docker, monitoring/alerting
• Other: Computer vision (RTSP/IP cameras, object detection), Speech-to-text (Indian languages)
• Infrastructure: Isolated VPS per client, queues, webhooks, retries, idempotency

Requirements

Must have:

• 4+ years of production experience with TypeScript / Node.js — building backend systems that real users depend on daily.
• Practical experience with WhatsApp automation libraries and session-based bridges (e.g., whatsapp-web.js, Baileys, Venom, or similar) — you have kept a session alive in production, not just run a demo.
• Solid grasp of queues, webhooks, retries, idempotency, and rate limits — systems that hold up when volume spikes.
• Hands-on work with LLM APIs and agent patterns: tool calling, structured outputs, context management, evaluation.
• Comfortable with Linux, VPS deployment, Docker, and day-to-day DevOps without hand-holding.
• Ability to spot a fragile system and fix it before a client notices.

Good to have:

• Meta / WhatsApp Business Platform (Cloud API) experience — BSP setup, template approvals, group and media handling.
• Computer vision in production: RTSP / IP camera streams, object detection, running models on-prem.
• Speech-to-text or evaluation work in Indian languages.
• Experience at an early-stage startup or on a small product team.

You will do well here if:

• You can read Hindi, Marathi, or Gujarati well enough to sanity-check what the model is doing.
• You are willing to occasionally visit a construction site to understand why a feature isn't working.
• You prefer owning a system properly over being one of thirty engineers on a large one.
• You work well remotely — clear written updates, responsive during IST hours, no chasing required.

• Work from home (fully remote)
• Small, focused team with immediate impact
• Opportunity to work across LLM agents, real-time messaging, and computer vision
• Paid take-home assignment, no unpaid multi-round marathons