REF92857P_2026224137 - AI/ML Engineer - 4 to 8 years - Pune/Vizag (WFO)

We are seeking a highly skilled Agentic AI Engineer to build and deploy multi-agent, goal-driven automation for document-heavy logistics workflows. The role owns the end-to-end agent lifecycle: from email/document ingestion to orchestrated workflow execution, system integrations (TMS/BL platforms), and a robust human-in-the-loop (HITL) + audit layer required for regulated shipping documentation.

This is not a “model-only” ML role. You will engineer production-grade agentic workflows where agents do the work, the orchestrator decides what runs, exceptions are routed correctly, and every action is traceable.

Key Responsibilities

1) Agentic Workflow Orchestration (Core)

Design and implement multi-agent architectures (classification, extraction, validation, customer follow-up, drafting, amendments, release) under a unified orchestrator that routes tasks, handles retries, manages state, and enforces guardrails.

Build case/task management for shipment documentation workflows: SLA prioritization, escalation rules, exception categories, and queue-based operations (shadow → assist → auto).

Implement confidence-driven automation (auto-run vs escalate vs stop) and structured fallbacks when upstream data or system access is limited.

2) Enterprise Integration (TMS / BL / eBL / Content Systems)

Build secure integrations to enterprise systems using REST/SOAP APIs where available; design pragmatic fallbacks (file-drop, staging UI, controlled automation) when direct APIs are constrained (e.g., Citrix-hosted systems).

Integrate with:Outlook/email ingestion and communication loops (request missing info, reminders, threaded responses).

Digiview / content repositories for archiving and retrieval of instruction/amendment emails and supporting documents.

BL platforms / eBL networks as required by process design (draft → review → release).

Create robust integration patterns: idempotency, deduplication, rate limiting, secure service accounts, sandbox/testing modes.

3) GenAI + RAG for SOP-grounded Reasoning

Implement LLM-powered capabilities for classification, extraction, SOP-grounded validations, and structured decision support using RAG (vector DB), prompt engineering, and context management.

Optimize token usage and response structure for cost-efficient, scalable throughput.

4) Document Intelligence & Data Pipelines

Build document handling pipelines (OCR/PDF parsing, table extraction, field normalization) for SI/draft/amendment content, including multilingual and semi-structured formats.

Engineer data pipelines to support continuous improvement: training data capture, labeling workflows, replay harness, and error analysis.

5) Human-in-the-Loop (HITL) Console, Audit & Controls

Build a HITL review/approval layer (draft BL review, exception resolution, amendment approvals) with role-based access controls and supervisor capabilities—treated as a peer system with its own logs and controls.

Implement a full audit trail: every automated/manual action logged with timesta