Overview

Position: AI Workflow Engineer

Responsibilities

• Build multi-agent systems on IBM Watsonx Orchestrate (or equivalent: LangGraph, CrewAI, AutoGen) — including routing logic, tool selection, anti-hallucination guardrails, and output formatting rules.
• Develop NL-to-SQL microservices over watsonx.data / Presto / Iceberg: schema grounding, query validation, SQL injection hardening, post-processing for sort/totaling/code substitution, and .env-based credential management.
• Ship RAG systems end-to-end: chunking strategy, embedding and retrieval (Milvus /
• Integrate enterprise systems via OpenAPI and OAuth2: SAP procurement/CRM datasets,
• Refactor monolithic prompt chains into modular, testable, multi-stage pipelines (e.g., a single 500-line prompt becoming a clean 4-stage pipeline with intermediate validation).
• Wrangle messy enterprise data: consolidate SAP fact tables, resolve data quality issues (float corruption on material codes, fiscal-year edge cases, negative-value preservation, etc.), and design clean schemas for downstream agents.
• Build FastAPI microservices with proper auth, request validation, error handling, logging,
• Author agent behavior configurations and runbooks: routing rules, fiscal year conventions, funnel display logic, anti-hallucination policies, and edge-case handling.
• Automate process classification — e.g., classifying hundreds of SOP steps for automation
• Document everything: architecture diagrams, ownership matrices, API contracts, and decision logs.

Qualifications

• Location: Delhi NCR (on-site)
• Team: Data & AI Engineering
• About the Role
• We're looking for an AI Workflow Engineer to design, build, and ship agentic AI systems and intelligent automation pipelines on top of enterprise data platforms. You'll be turning
• ambiguous business problems — procurement analytics, CRM insights, HR workflows,
• document Q&A — into production-grade AI workflows that real users depend on.
• This role sits at the intersection of LLM orchestration, data engineering, and enterprise
• integration. You won't just be writing prompts; you'll be designing multi-stage pipelines,
• hardening them for security, integrating them with SAP/Microsoft/Google systems, and
• operationalizing them on platforms like IBM watsonx.
• Python (2+ years): FastAPI, async/await, type hints, Pydantic, clean module structure, and ahealthy aversion to copy-paste code.
• SQL (intermediate to strong): CTEs, window functions, joins across 5+ tables, and the ability to debug a slow Presto/Postgres query without panicking.
• Production LLM experience: at least one shipped project using RAG, structured outputs,
• Prompt engineering fluency: few-shot design, chain-of-thought, structured JSON outputs, prompt injection awareness, and iterative evaluation.
• At least one orchestration framework: Watsonx Orchestrate, LangGraph, LangChain,
• Vector databases: Milvus, pgvector, Pinecone, Weaviate, or Chroma — comfortable with
• API integration: OAuth2 flows (auth code, client credentials, delegation), OpenAPI 3.0, REST debugging with curl/Postman.
• Git hygiene: clean commits, branch workflows, code review participation, and the ability to refactor without breaking things.
• Security mindset: SQL injection awareness, secret management via .env/vaults, input