Agentic AI Automation Engineer

HRS Group

Remote

Mid-Level AI Engineer — Agentic AI

AI-Workflow Programme | Mohali, On-Site

POSITION

We are seeking a Mid-Level AI Agentic Engineer to join the AI-Workflow programme and

build the autonomous crew systems that augment HRS operations across Finance,

Controlling, Operations, Customer Service, Customer Experience, and HR. This is not a

research role or a prototype environment — you will be building production AI crews that

handle live operational workflows for real departments, with real outcomes measured

from day one.

You will work within a "crews building crews" model: a platform of seven build agents (PM,

Architect, Automation, QA, SRE, Documentation, Observability) scaffolds, tests, and

documents the operational crews you build. Your job is to close the gap between agent

scaffolded output and production-ready code — working directly with the Tech Lead, the

PM, and the build agent platform to deliver tested, instrumented, and documented crews

within a 10-day delivery lifecycle.

The mission is workforce augmentation. AI handles the volume. Humans handle the

judgement. Every crew you build encodes that principle in every escalation boundary,

every guardrail, and every human-in-the-loop gate.

CHALLENGE

Crew Development & Implementation

Build operational AI crews from structured To-Be process descriptions using DSPy

typed signatures with assertion guards and agent workflow orchestration patterns

such as state machines, human-in-the-loop checkpoints, and resumable execution

Implement N8N workflow automation and JSON integration connectors linking

crews to operational systems including Zammad, Genesys, and enterprise back

office platforms

Work directly with the Automation Agent to scaffold DSPy modules and agent

workflows — extending and improving generated output, not accepting it verbatim

Design and encode specific, testable escalation boundaries for every crew before

shadow deployment — grounded in real process context, not generic confidence

thresholds

Deliver every crew with 100% unit test coverage, a complete runbook, and New

Relic instrumentation live before go-live — these are deployment gates, not

aspirational standards

Contribute reusable patterns to the shared crew library and peer-review modules

built by other engineers on the team

Technical Execution & Quality

Implement agent memory management using explicit typed state, structured

context handling, and clear handoff boundaries to prevent context degradation

across multi-step operational workflows

Apply three-layer output validation — DSPy assertions, output validators, and policy

enforcer — on every crew module before merge

Build and validate test suites covering non-deterministic edge cases and failure

modes — not just happy paths — using the QA Agent's generated baseline as a

starting point

Integrate crews with AWS Bedrock model routing (Claude Haiku/Sonnet) and work

within the EKS and Terraform IaC stack managed by DevOps

Maintain guardrails configuration for every crew — escalation triggers, human

approval gates, and policy enforcement — encoded in config before any crew

enters shadow deployment

Participate in weekly DSPy evaluation cycles against gold-standard baselines to

validate crew output quality and flag drift

Observability & Production Operations

Instrument every crew with New Relic metrics from day one: throughput, error rate,

latency, escalation rate, and cost per task — observability is a deployment

prerequisite, not an afterthought

Actively diagnose and resolve production failure modes: memory drift across multi

step workflows, hallucination under low-confidence RAG retrieval, context

degradation in long-running state machines, and prompt injection via untrusted

integration inputs

Use post-deployment observability data to identify improvement candidates and

raise them in RAID — closing the feedback loop the Observability Agent depends on

Contribute to the continuous improvement cycle: every crew in production is a

measurement and improvement loop, not a delivery milestone

Collaboration & Build Platform

Work within the 10-day delivery lifecycle — Request → Discovery → Design →

Development → QA → CI/CD → Monitoring → Continuous Improvement — delivering

to standard at each stage

Collaborate with the PM during Discovery to assess process automation feasibility

using FUDV scoring — frequency, uniformity, digitisation, volume — and push back

credibly where AI reliability or data quality is not there yet

Contribute to Thursday technical reviews and Friday retrospectives with substantive

input — not status updates but engineering judgment

Use the build agent platform as a personal productivity multiplier — flag platform

gaps via RAID rather than working around them silently

FOR THIS EXCITING MISSION YOU ARE EQUIPPED WITH…

Agentic AI Technical Skills

3–5 years of experience in AI/ML development with 1+ years in agentic AI or

advanced LLM applications shipped to a production environment — not prototype

or hackathon experience

Hands-on experience with DSPy typed signatures and assertion guards — not just

LangChain familiarity

Practical exposure to at least one agent framework or platform such as LangGraph,

Google ADK, Amazon Bedrock AgentCore, LangChain, CrewAI, or equivalent; the

role values transferable agentic engineering patterns over any single required

framework

Experience building and committing N8N workflow automation in a production

codebase

Demonstrated ability to design specific, testable escalation boundaries in a live

operational AI system

Can show their work — a GitHub profile, a shipped system, or a concrete

before/after on a workflow they automated carries more weight than academic

credentials

AI Engineering Capabilities

Strong Python programming skills with AI/ML libraries and practical agentic

engineering patterns; able to work across frameworks when needed, with exposure

to at least one of LangGraph, Google ADK, Amazon Bedrock AgentCore, LangChain,

CrewAI, or equivalent

Production experience with AWS Bedrock or equivalent cloud-based LLM routing

and model management

Knowledge of vector databases, embedding systems, and retrieval-augmented

generation — including retrieval quality assessment and hallucination mitigation

Understanding of MLOps and AIOps practices: CI/CD for AI systems, evaluation

harnesses, and gold-standard baseline testing

Hands-on experience with New Relic or equivalent observability tooling for

production AI systems — metric design, dashboard instrumentation, and anomaly

diagnosis

Familiarity with containerisation, EKS, and Terraform IaC sufficient to work within a

DevOps-managed infrastructure without creating integration delays

Development & Process Skills

Test-driven development for non-deterministic systems — 100% unit test coverage

before merge is a non-negotiable standard in this team

Experience with agile delivery in a timeboxed sprint model — able to take a

structured process description from design to shadow deployment within a 10-day

lifecycle

Strong code documentation discipline — every module peer-handoff ready, every

runbook complete during build, every decision traceable in RAID

Ability to work within an architecture set by a Tech Lead — executing with full

ownership and quality pride within defined guardrails, escalating cleanly when

constraints need revisiting

Professional Skills

Writes to be understood, not to be impressive — RAID entries a director can triage,

runbooks a department SME can follow, code a peer can extend without asking the

author

Calm under non-determinism — diagnoses production failures methodically using

observability data rather than thrashing or going silent

Dog-food mentality — uses the build agent platform to accelerate their own work

and actively contributes to improving it

Mission-driven — understands that the goal is workforce augmentation, not

automation for its own sake, and builds every crew with that principle at its centre

Preferred Experience

Domain experience in at least one of: Finance, Controlling, HR, Operations,

Customer Service, or Customer Experience — brings escalation boundary instinct

that no intake card can fully replicate

Experience with enterprise system integrations: Zammad, Genesys, UiPath, or

equivalent CRM/CX/RPA platforms

Familiarity with ChromaDB/Milvus/pgVector or equivalent vector store for RAG

pipeline development

Experience contributing to a shared pattern library or internal engineering

knowledge base in a multi-engineer AI team