Senior Generative AI Engineer — Agentic Systems

About the role

We build production agentic AI applications multi-agent systems that plan, execute, and self-correct on real customer data. Our agents run real ad campaigns and integrate with enterprise tools through MCP. They handle real money and real consequences, which means hallucinations, latency spikes, and broken tool calls are your problem, not a research curiosity.

We're hiring an engineer who has shipped non-trivial GenAI applications to production and wants to own the next generation of ours. This role is about building things people use, not training models from scratch. If you love turning LLM capabilities into reliable products, this is for you.

What you'll own

1. Agentic applications in production. Design, build, and operate multi-agent workflows (CrewAI, LangGraph, or equivalent) that run unattended against customer workloads analytics agents, campaign management agents, conversational interfaces over enterprise data.
2. End-to-end product features. Take a use case from "what would be useful?" through prompt design, tool wiring, eval harness, frontend integration, and rollout owning the full stack of a GenAI feature.
3. Non-chat AI experiences. Most of our surfaces are not chatbots they're dashboards with AI-generated insights, scheduled briefs, anomaly cards, recommendation panels, inline "explain this" affordances, and agent-triggered workflows that surface results in the UI. You'll design AI features that fit naturally into product UX, not just behind a text box.
4. Retrieval that works. Practical RAG hybrid retrieval, re-ranking, chunking strategy, retrieval evals, and the discipline to tell when retrieval is the bottleneck vs. when the model is.
5. Tool use and integrations. Build MCP servers, design function-calling contracts, and wire agents into messy real-world APIs (ad platforms, CRMs, warehouses, BI tools).
6. Cost, latency, and reliability. Token budgets, prompt caching, model routing across providers (Claude / GPT / open-weight), and graceful degradation when a provider is down.

What we expect you to bring

1. 5+ years total engineering experience, with 2+ years building GenAI applications that real users depend on. A single course project or hackathon RAG demo will not clear this bar.
2. At least one production GenAI application you can talk about in detail what it does for users, the architecture, what broke, what you measured, what you'd do differently. Be ready to whiteboard it.
3. Strong product instincts you can look at an LLM capability and figure out the right shape of feature to build around it, including when not to use a chat interface.
4. Deep comfort with the modern application stack: Python, CrewAI / LangGraph / LangChain / LlamaIndex, vector stores (pgvector, Pinecone, Weaviate), Claude & OpenAI APIs, MCP, prompt caching, structured outputs, tool calling, streaming.
5. Solid full-stack sensibility you understand how GenAI features land in a UI (Next.js / React or equivalent), including non-chat surfaces like dashboards, cards, inline suggestions, and async report generation. Frontend doesn't need to be your primary craft, but you should be comfortable shipping the full feature.
6. Experience designing structured agent outputs that downstream UI components can render JSON contracts, streaming updates, progress states not just freeform text into a chat bubble.
7. Experience with observability for LLM apps (Langfuse, Arize, Helicone, or homegrown) and the instinct to instrument before shipping.
8. Cloud production experience on AWS, GCP, or Azure Lambda / containers / queues, not just notebooks.
9. Strong written communication. We work async; clear design docs and incident write-ups matter.

Nice to have

1. Open-source contributions to agent frameworks, eval libraries, or MCP servers.
2. Experience with Adtech, Analytics, or other domains where agent decisions translate to dollars.
3. Fine-tuning experience (LoRA / QLoRA) and a point of view on when it's worth it vs. when prompt engineering and retrieval are enough.
4. Background in classical ML, data engineering, or distributed systems before pivoting to GenAI applications.

Not a fit if

1. Your GenAI experience is one RAG project from a bootcamp or a final-year universityproject.
2. You've never debugged a hallucination in production at 2 AM.
3. You're hoping the role will teach you LLMs from scratch.
4. You see yourself primarily as a researcher or model trainer rather than an application builder.
5. Your only mental model for "AI in a product" is a chatbot.

Compensation & logistics

Senior , commensurate with experience. Fully remote within India our team is Bengaluru-anchored, so expect collaboration on India business hours with occasional in-person meetups in Bengaluru. Equity for the right candidate.