Senior Full Stack Engineer — AI & Agentic Systems

We are looking for an exceptional Senior/Staff Full Stack Engineer with 10+ years of engineering experience to build the next generation of AI-native software systems.

This is not a traditional full-stack role. We are looking for someone who understands how to build systems where AI agents can reason, execute, observe outcomes, learn from feedback, and continuously improve their own performance.

You will work across the entire stack — from frontend experiences and APIs to distributed systems, data infrastructure, LLM orchestration, agentic workflows, evaluation systems, and autonomous optimization loops.

The ideal candidate has deep hands-on experience with Claude/Anthropic models, agentic architectures, tool-use, multi-step reasoning, autonomous execution loops, and AI-driven self-optimization.

You should be comfortable asking: \"How can we make the system improve itself rather than requiring an engineer to manually optimize every workflow?\"

What You'll Build

Systems capable of understanding complex business objectives, breaking them into executable tasks, selecting and using the right tools, executing multi-step workflows autonomously, observing results, evaluating whether the outcome was successful, identifying failures and inefficiencies, modifying strategies based on what it observes, running experiments, learning from historical execution data, optimizing future decisions automatically, and escalating to humans when confidence is low.

1. Agentic AI Architecture

Design and implement production-grade agentic systems using models such as Claude. Agent orchestration, tool calling, function calling, multi-agent architectures, planning and task decomposition, agent memory, context management, state machines and workflow engines, long-running agents, human-in-the-loop systems, autonomous execution, recovery and retry mechanisms, observability, and evaluation. You understand the difference between LLM → Agent → Workflow → Autonomous System, and when each is appropriate.

2. Agentic Loops & Self-Optimization

A major part of the role is building closed-loop systems: Goal → Plan → Execute → Observe → Evaluate → Learn → Re-plan → Execute. Systems that evaluate their own outputs, detect failed actions, identify root causes, adjust strategies, optimize prompts and tool selection, keep what works, roll back what does not, and improve over time. Reflection, critique, self-evaluation, feedback loops, reward signals, evaluation frameworks, automated experimentation, memory, retrieval, state management.

3. Claude / LLM Engineering

Deep practical experience with Claude/Anthropic APIs is highly desirable. Tool use, structured outputs, streaming, context management, prompt engineering, system prompts, long-context workflows, model routing, token optimization, latency and cost optimization, context compression, agent memory, LLM evaluation. Experience with other frontier models (OpenAI, Gemini, Llama or equivalent) is a plus.

4. Full Stack Engineering

Frontend: React, Next.js, TypeScript, modern component architectures, state management, real-time and streaming AI interfaces, agent activity and execution interfaces, data visualization.

Backend: Node.js, TypeScript, Python, REST APIs, GraphQL, WebSockets and streaming, event-driven architectures, background workers, job queues, distributed systems, authentication and authorization.

Design systems that reliably execute thousands or millions of AI and data-processing tasks. Kubernetes, Docker, Cloud Run and serverless, message queues, Redis, Kafka or equivalent, distributed job processing, concurrency management, rate limiting, retries, idempotency, fault tolerance, observability. You know how to build systems that stay reliable when agents fail, APIs time out, models hallucinate, or downstream services go away.

Build the infrastructure agents need to learn from historical executions. PostgreSQL, BigQuery or equivalent data warehouses, ClickHouse or analytical databases, vector databases, embeddings, retrieval systems, event logs, feature stores, analytics pipelines, data ingestion.

7. AI Evaluation & Observability

Build systems that objectively determine whether an agent is getting better. Automated evaluations, regression testing, LLM-as-a-judge, success/quality/cost/latency metrics, tool success rates, failure analysis, agent trajectory analysis, experimentation frameworks, A/B testing. Treat an AI agent like a production system that needs testing, measurement, debugging and continuous optimization.

8. Autonomous Optimization

Build mechanisms where the system discovers better ways to accomplish a task: automatically selecting the best model, the best tool sequence, optimizing prompts and workflows, learning which strategies produce better outcomes, detecting inefficient workflows, running controlled experiments, comparing execution trajectories, promoting what works and rejecting what does not.

Required Experience

AI / Agentic Systems: strong hands-on experience with several of — Claude/Anthropic, LLM APIs, agentic workflows, tool-using agents, function calling, multi-agent systems, agent memory, RAG, planning systems, reflection loops, evaluation frameworks, autonomous workflows, AI experimentation, LLM observability, prompt optimization, model routing.

What We Value

We are particularly interested in engineers who have built these systems themselves, not engineers who have only read about them. You should be able to show us production AI systems, agents executing real tasks, complex tool-use workflows, autonomous or semi-autonomous systems, AI evaluation systems, self-improving workflows, and large-scale distributed systems.

Ideal Candidate

You have the mindset of a Staff Engineer + AI Engineer + Distributed Systems Engineer. You enjoy problems where the architecture is not obvious and conventional software engineering is not enough. You move comfortably between frontend → backend → infrastructure → data → LLMs → agents → evaluation → optimization. You do not just ask \"how do we implement this?\" — you ask \"how do we design this so the system can eventually figure out the best way to implement and operate it itself?\"

Nice to Have

Seniority Expectations

This is a 10+ year engineering role. You own architecture rather than implement tickets. You make independent technical decisions, identify architectural weaknesses proactively, mentor engineers, establish engineering standards, debug complex production systems, think about scalability from day one, understand the trade-offs between performance, cost and reliability, translate ambiguous business problems into technical systems, prototype rapidly and productionize what works, and challenge assumptions when necessary.

First 6 Months

Establish the core agentic architecture · Build production-grade Claude integrations · Build reusable agent and tool infrastructure · Establish agent state and memory architecture · Build agent evaluation infrastructure · Implement execution tracing and observability · Build closed-loop execution systems · Introduce automated optimization experiments · Establish reliability and safety mechanisms · Help define the architecture for a self-optimizing AI platform.

Interview Process

Architecture

— design an autonomous agent that receives a business objective, decomposes it into tasks, executes tools, evaluates results, and retries or changes strategy when execution fails.

AI Systems

— design a system that lets an agent improve its performance over thousands of executions without blindly modifying production behaviour.

Engineering

— design a distributed execution system running thousands of concurrent agent workflows while handling rate limits, failures, retries and partial execution.

Practical

— build a small agent using Claude that plans → calls tools → observes results → evaluates itself → changes strategy → completes the task.

Compensation

RM 15,000 – RM 30,000 per month, based on experience, technical depth, and demonstrated ability to build production-grade AI systems. Exceptional candidates with significant experience building agentic or self-optimizing systems will be considered at the top of the range.

The Core Question

If you are excited by the idea of building software that does not just execute instructions, but can reason, act, measure outcomes, learn from failures, and continuously become better — we want to hear from you.