Scavenger Tech

We are seeking an elite, highly independent Principal Software Engineer & Architect who views advanced AI coding harnesses and multi-agent frameworks not as a crutch, but as a force multiplier for rapid, high-quality execution.

In this role, you will not be a cog in a massive machine, nor will you be stuck managing a single legacy codebase. Instead, you will act as a solo "special forces" engineer, moving dynamically from project to project. You will take loose, highly ambiguous business requirements and autonomously transform them into production-ready, enterprise-grade software. Your projects will span the entire functional spectrum—from building complex client portals to engineering mission-critical dispatch systems.

Because our applications deploy across a mix of Azure cloud and secure, on-premises environments, you must possess an uncompromising mindset regarding data sovereignty, rigid deployment constraints, and architectural security.

What You Will Do

• Architect & Execute: Autonomously design, code, and deploy secure, scalable applications from scratch using C#/.NET and JavaScript/Node.js. You own the entire pipeline from blueprint to production.
• Navigate Ambiguity: Translate vague, high-level business concepts into robust system architectures without requiring granular specifications or hand-holding.
• Adapt and Pivot: Shift gears fluidly between rapid-prototyping greenfield sprints (2–4 weeks) and deep-dive enterprise deployments (3–6 months).
• Own the Security Lifecycle: Ensure all codebases are rigorously hardened against the OWASP Top 10 and strictly adhere to secure, on-premises infrastructure requirements.
• Orchestrate AI Workflows: Integrate your own preferred multi-agent frameworks and AI toolchains to accelerate development velocity, while maintaining absolute human oversight of logic, edge cases, and architectural integrity.

Who You Are

• An Independent Thinker: You don't wait for a ticket to be fully groomed. You figure out what needs to be built, challenge assumptions when necessary, and deliver the right solution.
• An AI Realist: You understand that AI is prone to hallucinations and lazy logic. You know how to orchestrate multi-agent workflows to handle the heavy lifting, but your own architectural expertise is what ensures the system is stable, secure, and performant.
• A Full-Stack Enterprise Builder: You have deep, production-tested experience within both the Microsoft ecosystem (C#, .NET) and modern JavaScript/Node.js environments.
• A Security-First Engineer: You have a proven track record of architecting for strictly controlled, hybrid, or on-prem environments where security cannot be an afterthought.

Requirements

• 7+ years of experience in enterprise-level software engineering and hands-on system architecture.
• Mastery of C# / .NET and JavaScript / Node.js.
• Strong experience designing hybrid architectures spanning Azure cloud and on-premises infrastructure.
• Deep, practical understanding of OWASP Top 10 principles and secure coding practices.
• Proven experience leveraging advanced AI coding assistants, custom multi-agent frameworks, or LLM-orchestrated workflows to significantly compress delivery timelines.
• Exceptional ability to work entirely solo, owning projects from concept to deployment.

Why Join Us?

If you are tired of endless committee meetings, rigid corporate bureaucracy, and being locked into one mundane codebase for years, this is your playground. You will have the autonomy to choose your own AI workflow toolchain, the freedom of a fully remote position, and the challenge of building entirely different enterprise systems from scratch back-to-back.

We build Arabic and English chat and voice assistants for UAE government services. They answer questions from official documents with citations, and they complete real transactions: traffic fines, licence renewals, payments, identity verification, all through SOAP and REST integrations with government systems.

Our work spans two worlds. Some projects run in the cloud with hosted models and managed GPUs. Others run fully on-premise inside air-gapped government networks: local models on our own GPU servers, no internet, no external APIs. You will work across both, and you will ship the same quality in each.

We are a small team and we deliver fast. A full system typically goes from requirements to production in under 30 days. That pace is possible because each engineer owns their work end to end: read the documents, build the feature, integrate the services, test the complete journey in chat and voice, package the release, and support it live.

What you will build

• Chat applications: multi-step citizen journeys with intent detection, tool calling, confirmations, and clean handling of records like vehicles, fines, and receipts. English and Arabic, including right-to-left layouts.
• Voice: streaming speech recognition and synthesis, barge-in, interruption handling, and smooth handoff from voice to visual UI for forms, payments, and OTP.
• RAG pipelines: ingestion of PDF, DOCX, scanned and bilingual documents; hybrid retrieval; reranking; citations that point to the exact source; evaluation so we know retrieval actually works.
• Agentic workflows: model reasoning combined with deterministic controls, so the assistant completes real services and never invents fees, eligibility rules, or records.
• Model serving: hosted model APIs and cloud GPUs on some projects; open models served locally with vLLM or similar on others, including quantization, latency, and throughput tuning on government hardware.
• Integrations: SOAP 1.1/1.2, WSDL, XML, and REST services; UAE PASS and OAuth identity flows; payment gateways with callbacks, receipts, idempotency, and audit logs.
• Deployment: containers, systemd, reverse proxies, TLS. For air-gapped sites, complete offline packages: Python wheels, model weights, container images, migrations, and rollback, with nothing silently downloaded at install time.

How we work

This part matters as much as the technical list. You will often be the only engineer on a delivery, working directly from client documents and existing code.

• You start from the documentation and the codebase, not from a task-by-task checklist.
• You search the docs, code, logs, WSDLs, and captured traffic before asking for help, and you escalate only when something genuinely blocks you: a missing credential, an external contract, a business decision.
• You fix root causes, not symptoms, and you check whether the same defect exists in the paths nobody reported.
• You test your own work across the whole journey before calling it done: chat, voice, retrieval, integrations, UI, deployment, and the scenarios your change might have broken. Done means verified, not "the model answered correctly once."
• You give short, honest progress updates with realistic estimates, especially when something slips.
• You stay accountable for the outcome the citizen sees, not just the component you wrote.

What we need from you

• 4+ years of production Python: async services, APIs, background workers, real error handling.
• LLM applications you actually shipped: RAG, tool calling, or agents doing useful work in production.
• Experience with hosted model APIs, and either experience running open models on GPUs or the clear ability to learn it fast.
• Solid Linux: containers, networking, TLS, and debugging a server that has no internet access.
• Integration skill: you can take a WSDL or API spec, get a working integration, and diagnose it when the live service disagrees with the documentation.
• The independent working style described above, for real, not as an interview answer.
• Based in the UAE or able to relocate quickly; the role is primarily on-site.

Strong pluses (we will teach what you are missing)

• Arabic, especially Gulf dialect, and English-Arabic code-switching
• Streaming speech systems: ASR, TTS, WebSockets, WebRTC
• GPU serving and quantization (vLLM, TensorRT-LLM, or similar)
• UAE PASS, UAE government services, or payment gateway integrations
• Air-gapped packaging: offline repositories, checksums, SBOMs
• SQL Server or comparable enterprise databases