We're building the financial infrastructure Gulf businesses have never had, and we need someone who can make it think, without letting it guess.

Noor Al Qallam Group is building Pragma, a business operating system for Bahrain and GCC SMEs — accounting, payroll, and compliance built entirely from scratch, not bolted onto an existing platform. We're now extending that into an AI-first layer: a business owner should be able to describe what happened in their business in plain language and have it reflected correctly in a real double-entry ledger.

Why this role is genuinely hard, and why it's not just "wire up a chatbot":

Accounting software has almost no tolerance for being approximately right. Every transaction has to balance to the fils. Every entry needs an audit trail. Multi-tenant data has to stay completely isolated between clients.

Layering AI on top of that means the model can't just produce a plausible-sounding answer; its output has to be validated against real accounting invariants before anything is allowed to post. That's a fundamentally different problem from building a customer-support chatbot, and it needs someone who understands how these models actually behave, not just how to call an API.

What you'll do:

- Build and extend Pragma's accounting engine (ledger, invoicing, AR/AP, reconciliation) in TypeScript on Supabase/PostgreSQL
- Design the AI layer that turns natural-language input into structured, correct financial entries — including the validation and guardrail logic that checks AI-generated output against accounting rules before it ever touches the ledger




- Use techniques like retrieval-augmented generation and embeddings to ground the model in a specific business's real chart of accounts and transaction history, rather than relying on generic model knowledge
- Build evaluation frameworks to actually measure AI accuracy before any feature is trusted with real client data — this isn't optional polish; it's core to the job
- Work directly with our Lead Developer on architecture, data integrity, and security for a system handling real money and real compliance obligations

What we're looking for:

- Solid TypeScript/Node.js and PostgreSQL experience
- Real ML/AI background- not just prompt-engineering experience, but a working understanding of how LLMs behave, their failure modes, and how to constrain them (formal ML education, published work, strong personal projects, or equivalent hands-on experience, we care about depth, not the credential itself)
- Practical experience with RAG, embeddings/vector search, and evaluating model output for correctness, not just building features that assume the model is right
- Comfort with financial or transactional systems, where correctness matters more than speed of shipping
- Startup mindset, small team, high ownership, building something that doesn't fully exist yet

What we offer:

- Competitive compensation based on experience
- Real equity in an early-stage company
- Direct work with the founding technical team, not buried in a large engineering org
- The chance to build compliance-grade AI-powered financial software used by real businesses from day one

If you've ever wanted to build AI that has to be right , not just convincing, send us an email and tell us why you are right for this job or, for a quicker reply, do the test: https://dev.pragmatechnology.in/apply/poisoned

📌 Software Engineer (India)
🏢 Noor Al Qallam
📍 India