Coffeebeans.io - AI Engineer - Deep Agent Architectures

CoffeeBeans; Odisha; On-site; INR 1,500,000 - 2,100,000; Full time

Job summary

CoffeeBeans is seeking an AI/ML engineer in India to architect complex LangGraph-based agent workflows. You will implement Deep Agents atop LangGraph, enabling dynamic task planning, sub-agent spawning, and persistent graph state management.

The role emphasizes robust HITL interrupt mechanisms and secure state handling. Applicants should demonstrate hands-on proficiency with LangGraph, Deep Agents libraries, and associated guardrails tools, and experience with modern telemetry and retrieval

Qualifications

• Expert-level mastery of LangGraph for building complex, cyclic, and stateful agentic workflows.
• Hands-on experience implementing the Deep Agents open-source library built on top of LangGraph, leveraging its native architectural patterns.
• Advanced reasoning: dynamic task planning (to-do lists), isolated sub-agent spawning, persistent virtual file systems within the graph state.
• Mastery over LangGraph state machines, including context compression middleware and interrupt nodes for HITL approvals.

Skills

LangGraph mastery

Deep Agents harness

Advanced reasoning

HITL state management

Tools

LangGraph library

Deep Agents library

NeMo Guardrails

Telemetry tools (LangSmith/Arize)

Job description

• Frameworks : Expert-level mastery of LangGraph for building complex, cyclic, and stateful agentic workflows.
• Deep Agents Harness : Hands-on experience implementing the Deep Agents open-source library built on top of LangGraph, leveraging its native architectural patterns.
• Advanced Reasoning Patterns : Proven experience building agents with built-in dynamic task planning (to-do lists), isolated sub-agent spawning, and persistent virtual file systems within the graph state.
• State Management & Interruption : Mastery over LangGraph state machines, including context compression middleware and the use of interrupt nodes to manage asynchronous human-in-the-loop (HITL) approval checkpoints.

Technical Skills & Requirements

AI Agents & Deep Agent Architectures :

• Frameworks : Expert-level mastery of LangGraph for building complex, cyclic, and stateful agentic workflows.
• Deep Agents Harness : Hands-on experience implementing the Deep Agents open-source library built on top of LangGraph, leveraging its native architectural patterns.
• Advanced Reasoning Patterns : Proven experience building agents with built-in dynamic task planning (to-do lists), isolated sub-agent spawning, and persistent virtual file systems within the graph state.
• State Management & Interruption : Mastery over LangGraph state machines, including context compression middleware and the use of interrupt nodes to manage asynchronous human-in-the-loop (HITL) approval checkpoints.

LLMOps, Evaluation & Guardrails

• Evaluation Frameworks : Extensive experience with Ragas (Retrieval Augmented Generation Assessment) to programmatically benchmark faithfulness, context recall, and answer relevance.
• Guardrails & Alignment : Implementation of strict security and behavior firewalls using frameworks like NeMo Guardrails, Guardrails AI, or Llama Guard.
• LLM Telemetry : Experience mapping evaluation metrics into observability tools such as LangSmith, Phoenix, TruLens, or Arize.

Advanced Retrieval (RAG) & Tool Integration

• Vector Ecosystems : Experience with vector databases (e.g., Pinecone, Chroma, Milvus, Qdrant) and advanced retrieval techniques (hybrid search, cross-encoder re-ranking).
• Deterministic Tool Calling : Expert-level Python or TypeScript proficiency to build robust agent function-calling mechanisms, integrating seamlessly with enterprise REST APIs, webhooks, and SQL/NoSQL databases.

Production Deployment & Monitoring

• Infrastructure : Deep understanding of containerized deployment (Docker, Kubernetes) within enterprise cloud environments (AWS, Azure, or GCP).
• Auditability : Experience setting up structured logging and immutable telemetry pipelines (Datadog, Prometheus, ELK stack) for complete audit trails.

(ref:hirist.tech)