AI Agent Engineer

Poaition: AI Agent Engineer Reports to: Head of Engineering / CEO Works with: Lead Developer, Plant Management, HR Location: Pune, India Type: Full-time, permanent Experience: 3+ years software engineering, with hands-on LLM application work Purpose of the role: To own MetisEI's AI agent layer end-to-end. We are rolling out approximately 30 agents spanning production, supply chain, quality, HR and finance. Several are already live at CT Mexico. These agents send reports and instructions directly to line managers and plant staff, largely over WhatsApp so reliability is the job, not a nice-to-have. Key responsibilities: Design, build and maintain the agent layer production efficiency reporting, workforce and absence reporting, supply chain, quality and finance agents. Enforce the deterministic/LLM boundary. All arithmetic, thresholds, bucketing, ranking and fixed content must be computed and validated in code. The LLM is used for natural language rewriting of a pre-verified output only. Build and maintain automated test suites and evaluation harnesses for every agent. Agents must be regression-tested before any prompt or logic change reaches a live recipient. Build output validation that blocks a message from sending if it fails its rules, rather than relying on the model to comply. Configure and operate the WhatsApp Business Platform (Meta Cloud API) business account and number setup, message template submission and approval, webhook handling, delivery and read receipts, error and retry handling, opt-in management, rate limits and quality rating. Build scheduled delivery pipelines (daily and weekly cadences, timezone-correct per site) including PDF report generation and attachment. Work directly with plant management and HR to specify agent behaviour, then translate that into testable rules. Maintain a versioned content library (tips, templates, standard messaging) with verbatim-copy guarantees where required. Essential requirements: Strong Python. This role writes real software, not just prompts. Hands-on production experience with LLM APIs (Anthropic, OpenAI or equivalent) not experimental or personal-project only. Demonstrable experience building evaluation and testing frameworks for non-deterministic systems. Practical understanding of where LLMs fail and how to constrain them structured output, validation, guardrails, fallbacks. WhatsApp Business Platform / Meta Cloud API configuration experience, or equivalent messaging platform integration (Twilio, SMS gateways) with clear ability to pick up WhatsApp specifics. Webhook and asynchronous job/scheduling experience. Ability to work with non-technical stakeholders and convert loose requirements into precise, testable rules. Desirable: PDF and report generation. Retrieval-augmented generation and vector databases (Qdrant). Manufacturing or operations domain exposure. Spanish.