Developer AI workflow

Position: AI Workflow Engineer

Experience:1–2 years

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.

What You'll Do

 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 /

pgvector), reranking, citation grounding, and evaluation harnesses.

 Integrate enterprise systems via OpenAPI and OAuth2: SAP procurement/CRM datasets,

Microsoft Graph (Outlook, Teams), Google Workspace, custom connectors with hand-authored OpenAPI specs.

 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,

and rate limiting.

 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

feasibility using a hybrid of rule engines and LLMs.

 Document everything: architecture diagrams, ownership matrices, API contracts, and decision logs.

Must-Have Skills

 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,

function/tool calling, or agent orchestration — not just notebook experiments.

 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,

CrewAI, AutoGen, or similar.

 Vector databases: Milvus, pgvector, Pinecone, Weaviate, or Chroma — comfortable with

chunking trade-offs and retrieval evaluation.

 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

validation, and rate limiting.

Pay: ₹200,000.00 - ₹250,000.00 per year

Work Location: In person