Forward deployed engineer (Software Engineer)

Instrumental; Dubai; Hybrid; AED 469,000 - 692,000; Full time

Benefits offered by this job

Health insurance

Paid annual leave

Ticket

Job summary

Instrumental is hiring for a software engineer to build end-to-end automation within client operations. You will handle messy data sources, unstructured inputs, and integrate AI where it adds value, while keeping the system deterministic and measurable.

You’ll work mostly remotely with on-site sessions for critical launches and discovery, collaborating with operations teams to deliver scalable, reliable solutions across ERP/CRM ecosystems.

Qualifications

• 3+ years building production software relied on by real users.
• Ability to design tech that fits client operations.
• Experience with unstructured input: PDFs, OCR and parsing emails.
• Proven integration experience: APIs, OAuth, webhooks and retries.
• Experience deploying services with Docker, cloud, and CI/CD.

Responsibilities

• Build and own production automation systems end to end inside client operations.
• Work with messy client systems: ERPs, CRMs and vendor platforms with poor APIs.
• Decide where deterministic code vs model is used and prove the model is reliable.
• Develop evaluation harnesses to measure error rate on real data.
• Keep systems running post-launch with monitoring and drift control.
• Collaborate with operations teams to deliver usable, scalable solutions.

Skills

Python

TypeScript

APIs

Background jobs

Queues

Idempotency

SQL

OCR

LLM APIs

Docker

CI/CD

Cloud

Tools

Docker

Job description

Most AI companies fall into one of two traps. Either they promise to AI-ify everything and ship

something nobody can depend on, or they build solid software and bolt an AI feature on top as a

We do neither. We build automation systems for real operations, where the software owns the control flow and AI is used only where nothing else works: messy documents, unstructured input, judgment at a volume humans can't sustain.

Every AI step in our systems has a defined contract, a measured error rate, and somewhere

to go when it isn't sure. If we can't measure it, we don't ship it.

What you'll do

• Build and own production automation systems end to end, deployed inside client operations
• Work with the client's actual mess: ERPs, CRMs and industry platforms never built to talk to each other, vendor systems with no real API, undocumented databases, scanned documents, Arabic and English in the same file, and the exports and email attachments people use to bridge the gaps
• Decide where deterministic code handles something and where a model has to, then prove the model is right often enough to trust
• Build the evaluation harnesses that tell us our error rate on real data, not on a benchmark
• Keep systems running after launch. Monitor, catch drift, retune thresholds, handle the exception queue
• Sit with operations teams, watch how they actually work, and come back with something that survives contact with them

What we're looking for

• 3+ years building and shipping software that real users or a real business depended on
• You can sit with a client's operations team, work out how their business actually runs, and come back with a technical design. This is the part we care about most
• Strong Python or TypeScript, and you have built and deployed real services: APIs, background jobs, queues, scheduled work, retries and idempotency. Not just scripts
• Solid SQL and schema design. You can model a client's process into tables and query it without help
• Real integration experience: third-party APIs, OAuth and API key auth, webhooks, pagination, rate limits, and the failure handling that comes with all of them
• You have worked with unstructured input: PDFs, scanned documents, OCR, parsing email and messages
• You can deploy and run what you build: Docker, a cloud provider, CI, logs and alerts. When something breaks in production, you can find the cause and fix itYou have used LLM APIs in production, including structured outputs and function calling
• You can own a piece of client work end to end without being managed through it, and you know when to escal…
• You write clearly. Most of the work is remote, so specs, updates and handover notes carry a lot of weight
• Previous consulting or client-facing engineering experience is a plus

How we work

Mostly remote.

On-site is for the parts that don't work any other way: discovery sessions at the client, go-live, and the days when something is broken in a way that needs you in the room. Expect a few days a month, clustered around those moments rather than spread evenly. We'll also get together in person regularly as a team.

You'll have a lot of room. We agree what needs to happen and by when, then you get on with it. No daily check-ins, no tracker to feed. We review the things that matter and help whenever you ask. The team is young and growing, so what you own expands as fast as you want it to.

What this is not

• Not a research role. We don't train models
• Not a prompt engineering role. Prompts are a small part of a much larger system
• Not an agent-building role. Agents are rarely our first choice, and the job is much wider than that
• Not a role where you hand over a prototype and move on. You own what you build
• Start: Sooner is better
• Language: Arabic is a strong plus, not a requirement
• Process: Intro call + Technical conversation
• Benefits: Health insurance, Paid Annual leave and Ticket.