Senior AI Engineer — Agentic Systems & Cloud Platform

<h1 style="text-align: center;">Senior AI Engineer — Agentic Systems &amp; Cloud Platform</h1> <p style="text-align: center;"><strong>Automations · Full-time · In-Person (Islamabad, PK) · Hours 6pm - 2am PKT&nbsp;</strong></p> <p><strong>About Nysonian<br></strong>Nysonian builds the next generation of global lifestyle brands, shaping how people travel, move, and live. We go beyond creating great products to build experiences that elevate everyday life and empower people around the world.</p> <p>Our Fast-Growing Portfolio Includes:</p> <ul> <li>NOBL Travel — redefining modern travel with design, durability, and performance</li> <li>FLO Pilates — bringing Pilates into homes and wardrobes globally</li> </ul> <p>With $350M+ in revenue, 400+ teammates across 8 countries, and 1M+ customers worldwide, we are shaping the brands that will define the next decade.</p> <p><strong>Core Values: </strong>Winners’ Mindset | Speed with Purpose | Thoughtful Innovation | Genuineness | No Ego, Full Ownership</p> <p><strong>The Opportunity<br></strong>You'll own the design and production operation of our agentic AI systems end to end; the models, the orchestration, the tools they call, the guardrails around them, and the cloud infrastructure they run on. You'll also work inside our commerce platform and internal CRM, because an agent is only as good as the systems it can reach.&nbsp;</p> <p>This is a builder role with real architectural ownership, not a prompt-tweaking role.</p> <p><strong>What You'll Own<br></strong></p> <ul> <li>Design and ship multi-agent systems; orchestrator/worker patterns, routing, delegation, and human-in-the-loop escalation that handle live customer and operational workflows.</li> <li>Build and maintain the tool layer: MCP servers, tool-calling schemas, and integrations across Shopify (GraphQL), ShipHero, Gorgias, Klaviyo, telephony, and internal APIs.</li> <li>Run a multi-model stack. Choose the right model for each job, route between them on cost and latency, and evaluate new releases against ours in production conditions.</li> <li>Own guardrails and safety: hallucination and grounding checks, structured-output validation, prompt-injection defense, PII handling, least-privilege tool scopes, and hard limits on any action that touches money or customer data.</li> <li>Build the evaluation and observability layer; golden datasets, regression suites, LLM-as-judge scoring, tracing, and dashboards so we know a change is an improvement before it ships.</li> <li>Design and operate the RAG layer: chunking, embeddings, hybrid retrieval, vector store hygiene, and keeping the knowledge base honest as policies change.</li> <li>Build and extend production services on AWS and on our MERN stack, including the Nysonik CRM.</li> <li>Take systems from prototype to production and keep them there: CI/CD, prompt and config versioning, staged rollouts, retries and idempotency on tool calls, cost tracking, and on-call ownership of what you build.</li> <li>Make every autonomous action that spends money idempotent, recorded and reconcilable; a retry must never pay twice, and "what did the AI issue last month" has to be answerable from a table, not an estimate.</li> <li>Keep the stack running when a provider doesn't: rate limits, quota exhaustion, model deprecations and outright outages with failover, graceful degradation, and alerting that fires on the absence of expected activity, not just on errors.</li> </ul> <p><strong>Skills &amp; Qualifications</strong></p> <ul> <li><strong>Cloud &amp; Production Software Engineering</strong><strong>&nbsp;</strong></li> </ul> <p>4+ years building and operating production software, with real ownership of what happens after deploy.&nbsp; Strong AWS (EC2, Lambda, S3, RDS, IAM, VPC, queues). GPU workloads and self-hosted inference are a plus.&nbsp;</p> <p>Containers, CI/CD, infrastructure-as-code, logging and metrics (Prometheus/Grafana or equivalent).&nbsp;</p> <ul> <li><strong> </strong><strong>Production Agentic AI</strong></li> </ul> <p>You have shipped agentic systems that run in front of real users, not just in a notebook. Deep practical experience with tool calling and function calling, MCP, structured outputs and schema validation, and recovering gracefully when a model returns something malformed.&nbsp;</p> <p>Multi-agent orchestration patterns, state and memory management, context-window budgeting, and async/event-driven design.</p> <p>Evaluation discipline — you can describe how you measured whether an agent got better.&nbsp;</p> <p>Cost and latency engineering: caching, batching, model routing, token economics.&nbsp;</p> <ul> <li><strong> </strong><strong>LLM Engineering at Scale</strong></li> </ul> <p>Serious prompt engineering across large, versioned, multi-section system prompts. Guardrails, grounding, hallucination mitigation, and output filtering as engineering problems with tests, not vibes.&nbsp;</p> <p>LLM security: prompt injection, jailbreaks, data exfiltration through tools, secrets handling. OWASP LLM Top 10 familiarity.&nbsp;</p> <p>Fine-tuning and adaptation (LoRA/QLoRA, distillation, preference tuning) — or a clear-eyed sense of when not to bother.&nbsp;</p> <ul> <li><strong> </strong><strong>AI System Design</strong></li> </ul> <p>You can take a fuzzy business problem and produce an architecture: what's a model call, what's deterministic code, what's a tool, where a human belongs, and what happens when each part fails.&nbsp;</p> <p>Sound instinct for the deterministic/model boundary. A rule that can be a regex should not be a paragraph of prompt, and a judgement call should not be a rule.&nbsp;</p> <ul> <li><strong> </strong><strong>Reliability and Measurement Discipline</strong></li> </ul> <p>This is the part we care about most, because it's where AI systems in commerce actually hurt you. You treat a success status as a claim, not a fact. The expensive failures in this domain are silent: a tool that returns 200 and writes nothing, a token that was revoked last week, a workflow whose error handler swallows the error and reports "ok". You verify the side effect, not the exit code.&nbsp;</p> <p>You can size an experiment. You know why a 3-point movement in a daily metric is usually noise, what a matched-window baseline is, and roughly how many conversations per arm you need before a prompt change can be called an improvement. You've migrated a production system across model versions and can say how you proved behaviour didn't regress.&nbsp;</p> <p>Debugging discipline in a stack where the same input doesn't always produce the same output: tracing, replaying real traffic, and reproducing a failure before fixing it.&nbsp;</p> <ul> <li><strong> </strong><strong>Commerce &amp; Enterprise Production Systems</strong></li> </ul> <p>Hands-on MERN (MongoDB, Express, React, Node.js) in a production commerce or enterprise context; comparable to our Nysonik CRM. Comfortable with REST and GraphQL integrations, webhooks, queues, and third-party APIs that fail in interesting ways.&nbsp;</p> <ul> <li><strong> </strong><strong>Staying Current</strong></li> </ul> <p>You track model releases, benchmarks, and the research that matters, and you can tell</p> <p>the difference between a genuine capability shift and a launch post. We'll expect you to bring that to how we build.&nbsp;</p> <p><strong>Nice To Have</strong><strong>&nbsp;</strong></p> <ul> <li>Warehouse management systems (ShipHero or similar), 3PL operations, inventory and fulfillment logic.</li> <li>Ecommerce platform depth: Shopify, order lifecycle, RMA and returns, subscriptions, multi-store or multi-region setups.</li> <li>Support tooling (Gorgias, Zendesk) and CX metrics — resolution rate, WISMO, first-contact resolution.&nbsp;</li> <li>Workflow automation platforms (n8n, Temporal, or similar); data engineering — ETL, analytics pipelines, reporting.&nbsp;</li> <li>Voice/telephony AI (ASR, TTS, real-time pipelines).&nbsp;</li> <li>Self-hosted or small-model inference, and a view on when a fine-tuned small model beats a frontier API call.&nbsp;</li> <li>Writing — docs, internal write-ups, post-mortems. We keep decisions written down.</li> </ul> <p><strong>Why You'll Love Working at Nysonian</strong></p> <p><em>Culture</em></p> <ul> <li>We're founder-led and operate with speed, direct communication, and clear accountability</li> <li>We invest in tools and management practices that help colleagues do their best work</li> <li>Our products are used by customers globally</li> <li>AI is intentionally embedded in how we work, create, and scale</li> <li>Senior leaders are expected to create structure, make decisions, and stay close to execution</li> </ul> <p><em>Growth &amp; Development</em></p> <ul> <li>Competitive pay and meaningful opportunities for performance-based advancement</li> <li>Ownership of strategy, systems, execution, and team buildout within your function</li> <li>Opportunity to build and scale a meaningful part of the business across two consumer brands</li> <li>Scope and compensation growth tied to performance, role expansion, and measurable business impact</li> </ul> <p><em>We are proud to be an equal opportunity employer. All qualified applicants will receive consideration without regard to race, color, sex, religion, gender, marital status, national origin, genetics, disability, age, veteran status or other characteristics.</em>&nbsp;</p> <aside class="job-preview__other-job-details"></aside> <p>&nbsp;</p>