AI Software Engineer (Backend, Agentic, and Voice AI)

AI Engineer — Agentic, Voice & Messaging Agents

Steps AI is hiring an AI Engineer in Hyderabad to build the AI agents that talk to customers on the web, WhatsApp, Instagram, Messenger and the phone.

Location: Hyderabad, India, on-site

Job type: Full-time

Experience: 2+ years

Focus: Agentic AI, GenAI, real-time voice agents, Meta messaging-channel agents

About Steps AI

Steps AI is an agentic AI platform that powers customer-facing AI agents for e-commerce, SaaS, healthcare, real estate, EdTech and financial services. Our agents go live in under five minutes from a single URL. They learn a business from its website, documents, FAQs and product catalog, then work across web chat, WhatsApp, Instagram, Messenger and voice.

We are profitable, growing fast, and ship production AI to customers every week. Voice is our newest product area, and we are building it now.

The role

You will build the agents themselves: the runtime that reasons and calls tools, the voice pipeline that talks to callers in real time, and the channel integrations that carry conversations over WhatsApp, Instagram and Messenger. You will work directly with the founders, own the features you build end to end, and ship them to real customers within days of designing them.

What you will buildAgentic AI runtime

Agent graphs with state, conditional routing, tool binding, checkpointing and human-in-the-loop interrupts

Escalation to human agents, with clean context handoff and resume

Multi-provider LLM orchestration (OpenAI, Anthropic, Gemini, Azure OpenAI) with per-agent model choice, fallback chains and cost and latency trade-offs

A tool ecosystem: Shopify and WooCommerce, CRMs and helpdesks, calendars and email, logistics carriers, web search, and channel-specific actions; new providers reach production in days, not weeks

Memory: long-term memory across sessions and short-term state per thread, with clear contracts between them

Streaming: token-level delivery, intermediate events, interrupt and resume, and structured end-of-turn events

Guardrails for prompt injection, PII and off-topic requests

Real-time voice agents

Streaming speech-to-text, LLM and text-to-speech pipelines, plus speech-to-speech models (OpenAI Realtime, Gemini Live) where they fit

Natural turn-taking: voice activity detection, endpointing, barge-in and interruption handling

A measured voice-to-voice latency budget, with sub-second responses as the bar

Inbound and outbound calling over telephony providers (Twilio, Exotel, Plivo, SIP) and WebRTC

Outbound campaigns with scheduling, concurrency limits, retries, voicemail detection, and consent and calling-hour rules

Warm transfer to humans, call recording, transcripts, post-call summaries and call analytics

English, Hindi, Telugu and other Indian languages, including code-mixed speech

Meta channel agents (WhatsApp, Instagram, Messenger)

The WhatsApp Business Platform (Cloud API): Embedded Signup onboarding, approved message templates, interactive messages, catalogs and the 24-hour customer service window

WhatsApp Business calling, so the same agent can answer voice calls inside WhatsApp

Instagram messaging (DMs and comment-to-DM replies) and Messenger, including handoff between the AI agent and human inboxes

Webhook ingestion that verifies signatures, is idempotent, and survives retries, duplicates and out-of-order events

OAuth and encrypted token storage per customer, plus subscription and rate-limit management

One customer identity and conversation history across web, WhatsApp, Instagram, Messenger and voice

Knowledge and retrieval

Ingestion of websites, documents, FAQs and product catalogs: crawling, parsing, OCR and layout-aware extraction

Hybrid retrieval (dense plus sparse) with cross-encoder reranking, tuned for recall and precision

Idempotent embedding and indexing pipelines on scheduled, long-running workflows

Platform, reliability and quality

Backend APIs in Python (FastAPI) and TypeScript (NestJS) for agents, workspaces, channels, integrations and billing

Multi-tenant workspaces with role-based access, quotas and usage metering

New modules designed end to end: data model, API, workflow, AI tool and UI

Contributions to the React/Next.js dashboard and the embeddable chat and voice widget

Instrumentation from day one: token usage, latency, answer quality, call quality, conversion and errors

Short design docs that explain the trade-offs behind what you build

Must-have skills

2+ years of professional experience building backend or full-stack applications

At least one LLM agent you shipped to real users (chat, voice or messaging) that you can demo live and walk us through

Strong Python: FastAPI, async I/O, Pydantic, structured logging

Production experience with LangGraph, LangChain or a comparable agent runtime: tool calling, state, checkpointing, streaming, human-in-the-loop

Direct integration with LLM APIs (OpenAI, Anthropic, Gemini or Azure OpenAI), including fallbacks and cost control

A vector database (Pinecone, Qdrant, Milvus or Weaviate), with hybrid retrieval and reranking

PostgreSQL schema design and query tuning, REST and webhook API design, and Redis or a job queue

WebSockets or other streaming protocols

Git, Docker, CI/CD, and deploying to AWS or GCP

Strong plus

Production voice agents: LiveKit or Pipecat; Deepgram, Whisper, AssemblyAI or Sarvam for speech-to-text; ElevenLabs, Cartesia or Sarvam for text-to-speech; Twilio, Exotel, Plivo or SIP for telephony; WebRTC and audio streaming

Meta platforms at scale: WhatsApp Cloud API, Embedded Signup, message templates and WhatsApp calling; Instagram and Messenger messaging; signed webhooks

TypeScript and NestJS with an ORM such as Prisma

React and Next.js, Tailwind, shadcn/ui or Radix, TanStack Query, and embeddable JavaScript widgets

Workflow orchestration: Temporal, Celery, Airflow or Prefect

E-commerce, CRM and helpdesk APIs: Shopify, WooCommerce, HubSpot, Zoho, Freshdesk, Zendesk

LLM observability and evaluation: Langfuse, OpenTelemetry, LLM-as-judge, RAGAS, red-teaming

Multimodal AI: vision-language models and live audio streams

Multilingual and Indian-language speech and text

Usage-based or subscription billing

Open-source work, papers, talks or hackathon wins in GenAI

How you work

You ship fast. A well-defined feature goes from idea to production in days, not weeks.

You read code faster than docs, and you trace a bug across services to its root cause yourself.

You design for failure: idempotent webhooks, correctly retried jobs, circuit breakers on external APIs, and guardrails on every prompt.

You measure what you ship: token usage, latency, retrieval quality, call quality and conversion, from day one.

You take ownership. When something breaks, you fix it, write it up, and prevent it from recurring.

You turn fuzzy business requirements into working features without rounds of hand-holding.

You are excited to build voice from its early stages.

Qualifications

B.Tech, M.Tech or MS in Computer Science, Software Engineering, AI/ML or a related field, or equivalent practical experience

2+ years of professional engineering experience with a production track record

Able to work on-site in Hyderabad, in person with the team

What we offer

Direct, no-gatekeeper access to the founders

Real ownership of what you build, including architecture choices within your scope

A fast learning curve on production agentic, voice and messaging AI

Competitive salary, calibrated to a live demonstration of what you can build

How to apply

Complete the application form. It takes about 5 minutes, and resumes sent outside the form are not reviewed. Shortlisted candidates hear from us within 48 hours. Come ready to demo, live, an agent you built.

Steps AI · stepsai.co · Hyderabad, India