Lead Engineer – Agentic AI

Qualminds; Hyderabad; On-site; INR 4,500,000 - 7,000,000; Full time

Job summary

QualMinds in Hyderabad is seeking a Senior Engineer to design and ship multi-agent AI systems for enterprise-scale applications. You will own the entire agent lifecycle, from orchestration to observability, and lead teams of specialized agents across planning, retrieval, and action execution.

The role requires 8+ years of experience, including 2+ years in production LLM/agentic systems with multi-agent architecture expertise, and strong Python/TypeScript skills for backend and frontend work.

Qualifications

• Bachelor's degree in Computer Science, Engineering, or equivalent practical experience.
• Proven experience designing and shipping multi-agent architectures: supervisor/orchestrator patterns, hierarchical and peer-to-peer topologies, planner–executor–critic loops.

Responsibilities

• Architect multi-agent systems end-to-end across lifecycle from scoping to deployment and observability.
• Design and build customer-facing voice + chat agent teams covering inventory search, finance estimation, trade-in valuation, offer discovery and scheduling.
• Ground all agent responses in live business data and implement guardrails and human-in-the-loop approval flows.

Skills

Multi-agent architecture

Python

TypeScript

Frontend/Backend

LangGraph

OpenAI Agents SDK

GM/Anthropic/Google SDKs

Realtime Voice AI

RAG & context engineering

Observability

Education

Bachelor's degree in Computer Science/Engineering

Tools

LangGraph

OpenAI Agents SDK

Google ADK

Anthropic Claude Agent SDK

Pydantic AI

Job description

At QualMinds, we design and develop world-class digital products and custom software solutions for Startups, Scale-ups, and Enterprises. Our engineering teams specialize in Frontend, Backend, Full Stack Development, QA Automation, UI/UX, DevOps, Data Science, Machine Learning, and Agile Delivery. We are passionate about building customer-centric software with the highest standards of quality, scalability, security, performance, and reliability.

Experience: 8+ years (2+ years in production LLM/agentic systems, with proven multi-agent architecture experience)

Job Summary

We are building multi-agent AI systems for the automotive dealership ecosystem — networks of specialized, collaborating agents spanning customer-facing conversational experiences (voice and chat agents on dealership websites that handle inventory search, payment estimation, trade-in valuation, offers, test drive and service scheduling, and lead capture) and internal business agents (conversational and autonomous operation of campaign management, lead management, offers, payments, reporting, forecasting, and email/SMS marketing — with proactive insights, recommendations, and human-approved actions).

These are not single-chatbot solutions: you will architect orchestrated teams of agents — planners, retrievers, domain specialists, action executors, and reviewer/critic agents — that coordinate, hand off, and share context to complete complex business workflows reliably. You will own these systems across the entire agentic development lifecycle: use-case scoping, multi-agent architecture and tool design, orchestration, evaluation, guardrails, deployment, observability, and continuous improvement.

Technical Skills and Experience

• Bachelor's degree in Computer Science, Engineering, or equivalent practical experience.
• Proven experience designing and shipping multi-agent architectures: supervisor/orchestrator patterns, hierarchical and peer-to-peer agent topologies, planner–executor–critic loops, agent handoffs, task decomposition and delegation, shared state/memory design, and failure isolation between agents.
• Hands-on with current agent frameworks and SDKs with strong multi-agent support: LangGraph, OpenAI Agents SDK, Google ADK, Anthropic Claude Agent SDK, Microsoft Agent Framework, Pydantic AI, or Vercel AI SDK — including subagents, handoffs, and graph-based orchestration.
• Deep experience with frontier model APIs: OpenAI (GPT-5.x, Realtime API), Anthropic (Claude Opus/Sonnet 4.x), Google (Gemini 3.x, Gemini Live); familiarity with open-weight models (Llama, Qwen, DeepSeek, Mistral) for per-agent cost/latency routing (e.g., small fast models for routing/critic agents, frontier models for planning).
• Expert in tool/function calling, structured outputs, MCP (Model Context Protocol) — designing and building MCP servers that expose business systems (inventory, CRM, campaigns, payments, reporting) as governed agent tools; working knowledge of A2A (Agent2Agent) for cross-platform agent interoperability.
• Realtime voice AI: speech-to-speech pipelines (OpenAI Realtime API, Gemini Live, LiveKit Agents, Pipecat), STT/TTS providers (Deepgram, ElevenLabs, Cartesia), barge-in/interruption handling, sub-second latency optimization, WebRTC and telephony integration — including voice agents that delegate to backend specialist agents mid-conversation.
• Production RAG and context engineering: vector stores (pgvector, Qdrant, Vertex AI Search, Pinecone), hybrid retrieval and reranking, context window management across agent hops, shared and per-agent memory (short-term, long-term, episodic), and freshness strategies for fast-changing data (inventory, offers, pricing).
• Agent evaluation and observability: LangSmith, Braintrust, Arize Phoenix, Langfuse, or custom harnesses — golden datasets, LLM-as-judge, multi-agent trajectory evals (correct delegation, handoff quality, loop/runaway detection), eval-gated CI/CD; OpenTelemetry GenAI semantic conventions with distributed tracing across agent boundaries.
• Guardrails and AI safety: prompt-injection defense, jailbreak resistance, PII redaction, content moderation, inter-agent trust boundaries, human-in-the-loop approval flows for consequential actions (payments, campaign sends, customer messaging), and compliance-aware messaging (TCPA).
• Strong Python and TypeScript; modern backend (FastAPI, Node.js) and frontend (React/Next.js) development, including streaming and realtime UX (SSE, WebSockets, WebRTC) and generative UI patterns.
• Cloud-native delivery: GCP (Vertex AI, GKE, Pub/Sub, Cloud Run, IAM, Secret Manager) or equivalent on AWS/Azure; Docker, Kubernetes, Terraform; durable execution for long-running agent workflows (e.g., Temporal) is a plus.
• Modern engineering workflow: GitHub (Actions, PR reviews), Jira/Linear, prompt and agent versioning, feature-flagged and eval-gated releases.
• Effective daily use of AI-assisted development tools (GitHub Copilot, coding agents, Claude Code, Cursor) and the judgment to review their output critically.
• PostgreSQL and modern data tooling; event-driven architectures (Pub/Sub, Kafka) for agent-triggered and agent-to-agent workflows.
• Strong written and verbal communication, including the ability to present AI risks, trade-offs, and recommendations to senior stakeholders.

Responsibilities

• Architect multi-agent systems end-to-end: decompose business workflows into specialized agents (routing, retrieval, domain reasoning, action execution, review/critique), choose the right orchestration topology per use case, and define agent contracts, handoff protocols, and shared memory/state.
• Design and build customer-facing voice + chat agent teams covering the full public website capability surface: inventory search and comparison, finance/payment estimation, trade-in valuation, offer discovery, appointment scheduling, and lead capture/qualification — with low-latency realtime voice, streaming chat, graceful fallback, and human handoff.
• Ground all agent responses in live business data with strict accuracy guarantees — no hallucinated prices, VINs, or offers — and implement brand-safe guardrails for public surfaces (prompt injection, jailbreaks, off-topic containment, PII handling).
• Build internal multi-agent workflows that operate backend business platforms conversationally and autonomously: campaigns, lead management, offers, payments, reporting, forecasting, and email/SMS marketing — following the insight → recommendation → approved action pattern with specialist agents collaborating under an orchestrator.
• Expose business capabilities as MCP tools with role-based permissions, audit logging, and approval policies scaled to action risk (read vs. draft vs. send vs. pay); enforce least-privilege tool access per agent.
• Build natural-language analytics: NL-to-SQL/semantic-layer querying, automated forecasting summaries, and scheduled executive digests.
• Own evaluation end-to-end: golden datasets per use case, offline regression suites, multi-agent trajectory and delegation evals, eval gates in CI/CD, A/B testing, and production quality monitoring (containment, conversion, CSAT).
• Establish LLMOps/AgentOps practices: prompt and agent versioning, per-agent model routing and semantic caching for cost control, token/latency budgets across agent chains, and observability dashboards with cross-agent distributed tracing.
• Lead root-cause analysis for agent failures (bad tool calls, wrong delegation, runaway loops, hallucinations, unsafe outputs) with durable systemic fixes and new guardrails.
• Partner with Product Owners, Architects, and platform teams to translate business workflows into agent capabilities and clear delivery plans; mentor engineers on multi-agent patterns, context engineering, and evaluation discipline; report AI system health, risks, and trends to leadership with supporting data.

Required Skills

LLMs Python Realtime Voice AI Multi-Agent Architecture MCP RAG AI Evaluation LangGraph