Requirements
- Experience building with large language models, including at least one workflow that ran against real business data.
- Practical command of tool and function calling: schemas, retries, partial failure, and what to do when a model calls the wrong thing.
- Python or TypeScript, and comfort with the queues, webhooks, APIs and cron jobs an automated workflow sits between.
- Context engineering: getting the right records, documents and business rules in front of a step, and keeping everything else out.
- A habit of designing the failure path first: what happens when a step is wrong, who is told, and how it is undone.
- Judgement about autonomy: which decisions a workflow may make alone, which need a person, and the ability to defend that line to a client.
- Able to explain a workflows behaviour and its limits to an operations manager who will be accountable for it.
Responsibilities
- Understand a clients existing process as it is actually performed, including the spreadsheets and the messaging groups, which is usually where the real process lives.
- Identify where AI can safely add value, and where automation without a model is the better answer.
- Design the workflow end to end: the steps, the tools, the data each one sees, and the points where it stops.
- Decide what runs unattended and what requires human approval, and build the approval step properly.
- Build the prompts, context and guardrails each step depends on.
- Connect workflows to business systems (ERP, CRM, commerce platforms, document stores) through APIs and webhooks, writing back through contracted interfaces rather than straight into somebodys database.
- Build evaluation and observability for whole workflows: end-to-end traces, graded outcomes, and a regression set that runs before every change.
- Work with our engineers to put workflows into production applications and keep them running.
- Measure the business effect (time saved, exceptions handled, error rate) and retire a workflow that is not earning its place.
Why this role matters
A model that answers well is a feature. A model that reads what arrived, decides what it is, does something in a system of record and knows when to stop is an operation, and that is where the value and the risk both are.
This role sits at the intersection of AI, software engineering, automation and business operations. You need all four to design a workflow a client will let run on a Monday morning without watching it.
What you ll work on
- Business process automation. Intake queues triaged and routed before a person opens them; repetitive back-office steps performed with an audit trail.
- CRM and ERP integrations. Workflows that read and write through documented interfaces to systems like Dynamics and Navision, because a system of record does not get replaced just because its interface looks old.
- eCommerce operations. Order exceptions, catalogue enrichment, supplier documents and returns handling inside the commerce platforms our clients trade on.
- Document workflows. Multi-step extraction, validation against business rules, and escalation of anything that fails a check.
- Customer support. Classification, drafting and routing, with the reply a person sends still a person s decision.
- Data processing. Reconciliation and enrichment where the exception matters more than the happy path.
What you ll bring beyond the requirements
- Production experience with an orchestration or agent framework, and a clear view of where it helped and where it got in the way.
- A background in workflow automation, integration or RPA before language models. Those instincts transfer directly.
- Familiarity with human-in-the-loop design: queues, approvals, and interfaces an operator will actually use.
- Observability practice: traces and dashboards that explain a run in business terms.
How we work
Every workflow we build is handed over: the client s repository, the client s accounts, and documentation good enough to run it without us. An automation nobody can inspect is one nobody will keep, so instrumentation and a plain account of what the workflow does are part of the deliverable.
You will work from the Chennai office on a hybrid basis, with delivery teams across Irving, Muscat and Chennai.