AI Data Engineer (Bangalore)

👋 About Xanadu

Xanadu helps mid-market organisations in the Middle East move from AI experiments to AI that actually runs in production. We’re a Certified Services Partner, Claude Partner Network, building on Claude for the reasoning, workflow and agent layer, and a Canva partner for the creative and communications layer.

We work in pods: a Forward Deployed Engineer onsite with the customer in Dubai, a Senior AI Engineer leading the build, an AI Engineer shipping it, and you owning the data that makes it work. An Associate AI Engineer works across both pods on testing and evals.

🗄️ The role

Agents are only as good as the data they can reach. In the mid-market, that data lives in SharePoint sites nobody has tidied, ERPs, CRMs, shared drives full of PDFs, and the spreadsheet someone emails round every Monday. You find it, make sense of it, get it flowing safely, and prove it’s good enough to trust.

The line in a pod is simple: you own everything up to the point where the agent asks for information. The engineers own what the agent does with it.

🎯 What you’ll own

• Data readiness. In discovery, work with the FDE to map where the customer’s data lives, how good it is and what it’ll take to use. Put the gaps in writing before anyone promises an outcome.
• Pipelines and ingestion. Design and run ETL/ELT from systems of record, with batch, incremental and CDC loads, orchestration and idempotent reruns. Equally, know when not to move data and let agents query the source live through APIs or MCP.
• Documents and retrieval. Parsing, chunking, metadata and retrieval for messy documents, including scanned PDFs and untidy SharePoint.
• Look-up tools. Build the tools agents use to find things in the customer’s data. The AI Engineer uses them; the pod’s Senior agrees how they fit together.
• Retrieval evals. Own the tests that answer one question: did the agent find the right information?
• Data handling. Access controls, least privilege, retention and residency, built to the pod’s single approach to who sees what (set by the Senior), so the customer’s risk team signs off. Gulf data residency and security rules are a hard requirement.
• Data quality. Checks that catch drift and breakage before users do.
• Test data. Realistic, safe datasets for building and testing, with the Associate AI Engineer.
• Patterns. Reusable connectors, ingestion and retrieval patterns into our Agent Library.

📏 What good looks like

• 90 days: Claude-certified, a data readiness assessment delivered on a live pod, first pipeline in production.
• 12 months: data foundations under two or more production deployments, and at least two reusable connectors or retrieval patterns in the Agent Library.

🔑 Must-haves

• 3+ years in data engineering, analytics engineering, or backend work that’s heavy on data.
• Strong SQL and Python.
• Hands-on ETL/ELT experience: orchestration (Airflow, Dagster or similar), transformation (dbt or similar), and managed or custom connectors.
• You’ve built and run production pipelines on a major cloud (AWS, Azure or GCP).
• Some hands-on experience with retrieval for LLM applications (embeddings, vector search or hybrid search), even if it’s from a project rather than production.
• Comfortable with enterprise data access: APIs, connectors, permissions, and systems that were never built to be integrated.
• Clear written English, and you can explain a data problem to someone who doesn’t work with data.

🌟 Nice-to-haves

• Strongly preferred: document retrieval for an LLM application running in production with real users.
• Search systems such as Elasticsearch or OpenSearch.
• Arabic, or right-to-left and multilingual data.
• Microsoft 365 and Graph, SharePoint, Salesforce, HubSpot or ERP data models.
• OCR and scanned-document pipelines.
• Data governance or privacy work, including regional residency requirements.
• MCP and the Claude Agent SDK.
• Real estate, financial services, or QSR and franchise data.

🚫 What this isn’t

• Not a BI or dashboard role. We’re not building reports; we’re feeding agents.
• Not a data science role. We don’t train models; we make them useful.

🧪 How we hire

1. Intro call (30 min).
2. Values and behaviours (45 min).
3. Bar-raiser (60 min).
4. Working session (90 min): we give you a messy set of documents and a question an agent needs to answer. Get the data into shape, build a simple retrieval step, and show us how you’d prove it works.
5. Decision: within 48 hours of every stage.

💛 Why Xanadu

• Every engagement is a new, real data estate. No toy datasets.
• Direct line to the FDE and the customer. You’ll know why the work matters.
• Certified on Claude on our dime, and a clear path to Senior.
• Equity (ESOP), so you share in what you build.
• No egos, no passengers.