Agentic Ai Engineer

Description:

Agentic Superintelligence Labs is a US-based AI company hiring immediately. This is a full-time, onsite role based in Karachi with a highly market-competitive compensation package. We are not looking for someone who builds demos. We need an engineer who has shipped production-grade autonomous AI systems that run continuously, recover from failures deterministically, and operate under real uptime and reliability consequences.

ABOUT THE ROLE

You will architect, build, scale, secure, and operate production-grade agentic systems from first principles — with zero fragility tolerance and no hand-holding. You will work across multi-agent orchestration, distributed cognition, AI infrastructure, long-horizon planning, real-time decision-making, event-driven execution, and cognitive runtime architecture. You will own everything end-to-end: architecture, orchestration, infrastructure, deployment, observability, security, and evaluation.

CORE TECHNICAL COMPETENCIES

Autonomous Agent Architecture Planner, executor, supervisor, worker, and critic systems. Hierarchical multi-agent orchestration, swarm coordination, reflection and self-correction loops, dynamic routing, state transition design, infinite-loop prevention, and human escalation architecture. LangGraph expertise for stateful, durable, long-running workflows is required.

Distributed Systems Engineering Fault tolerance, high-availability architecture, distributed coordination and consensus, graceful degradation, load shedding, backpressure handling, tail-latency optimization, deterministic replayability, rollback systems, idempotency, durable execution, SLO/SLA engineering, and event-driven architectures.

AI Infrastructure and ML Systems Distributed inference systems, multi-model orchestration, cost-aware and latency-aware model routing, batch vs streaming inference, multi-GPU serving, continuous evaluation, and ML observability. Stack includes vLLM, TensorRT-LLM, Ray, Kubeflow, and MLflow.

Reasoning and Cognitive Architecture ReAct, Chain-of-Thought, Tree-of-Thought, deliberative inference, long-horizon planning, self-reflection, critic models, goal decomposition, constraint satisfaction, PDDL, symbolic planners, policy enforcement, and autonomy boundary design.

Memory, State, and Persistence Episodic, semantic, working, shared, and task memory systems. Checkpointing, durable execution, replayability, memory compression, context pruning, and retrieval ranking. Databases: Postgres, Redis, pgvector, Pinecone, Weaviate, Chroma, Neo4j.

Tool Use and Enterprise Orchestration Tool schema design, function-calling optimization, parallel tool execution, dynamic tool discovery, tool reliability scoring, MCP integration, and production integrations with SAP, Oracle, Salesforce, ServiceNow, and Workday.

Multi-Agent Coordination Agent-to-agent protocols, message passing, consensus, negotiation, conflict resolution, task arbitration, blackboard architectures, A2A event bus design using Redis Streams and NATS, and deadlock prevention.

Context Engineering Prompt layering, dynamic context injection, attention budgeting, semantic compression, retrieval ranking, structured output, JSON mode mastery, and cost-aware multi-model routing across planner, executor, and fast model tiers.

Event-Driven and Async Architecture Kafka, RabbitMQ, AWS SQS/SNS, EventBridge, streaming events, dead-letter queues, pub/sub systems, Celery, Temporal, and Prefect for durable workflow orchestration, and full event-to-agent-to-tool-to-database-to-notification pipelines.

TECHNOLOGY STACK

Orchestration: LangGraph · CrewAI · AutoGen · Temporal · Prefect · Airflow · Dagster LLM Layer: LangChain · LlamaIndex · Haystack · OpenAI API · Anthropic API ML Infrastructure: Ray · Kubeflow · MLflow · vLLM · TensorRT-LLM Retrieval and RAG: pgvector · Pinecone · Weaviate · Chroma · Neo4j · Graph RAG Memory and State: Postgres · Redis · S3 · SQLAlchemy · Alembic Distributed Systems: Kafka · RabbitMQ · SQS/SNS · EventBridge · Redis Streams · NATS Evaluation: LangSmith · Ragas · DeepEval · Custom harnesses Guardrails: Guardrails AI · NeMo Guardrails · Policy checks · Sandboxing Runtime and API: FastAPI · Pydantic · asyncio · gRPC · GraphQL · Webhooks Deployment: Docker · Kubernetes · Helm · Terraform · GitOps · GitHub Actions Cloud: AWS · Azure · GCP · IAM/RBAC · Serverless Observability: OpenTelemetry · Datadog · Prometheus · Grafana · ELK Stack Security: OAuth/JWT · Vault · AWS Secrets · Azure Key Vault · RBAC Languages: Python — deep async mastery required

HARD REQUIREMENTS — EVERY ONE MUST BE TRUE

✅ You have built and operated autonomous or multi-agent systems in real production — not prototypes or demos ✅ You independently own architecture, orchestration, infra, deployment, observability, security, and evaluation end-to-end — no delegating ✅ You think in distributed systems, reliability engineering, and execution guarantees before writing a line of code ✅ You design full autonomous architectures independently, from first principles, without guidance or a technical lead ✅ You build security-first — prompt injection, privilege escalation, and audit requirements are architectural inputs, never afterthoughts ✅ You build evaluation systems before scaling — you measure task success, planning quality, tool correctness, hallucination rate, and cost efficiency ✅ You have built autonomous workflows that execute over hours or days — with durable state, mid-execution failure recovery, and deterministic replayability ✅ You have deep async Python mastery — asyncio, context managers, generators, concurrency, queues, dependency injection ✅ You reason quantitatively about latency, reliability, throughput, and cost tradeoffs — you set SLO targets and hit them ✅ You have integrated at least two of Salesforce, SAP, ServiceNow, Oracle, or Workday in a live agentic pipeline ✅ You architect escalation paths, confidence thresholds, interruptibility, and explainability from first principles