MLOps / LLMOps Engineer (Mid-Level)

Irth Solutions · Remote

About Irth Solutions

Irth Solutions is a leading provider of cloud-based SaaS software for damage prevention, asset integrity, stakeholder engagement and land management, helping energy, utility, telecom, and infrastructure companies protect their critical network infrastructure. With nearly three decades of industry experience, Irth serves customers across North America and continues to expand its platform with new data-driven and AI-powered capabilities.

MLOps / LLMOps Engineer – Insights (AI/ML)

Location: Remote – India
Department: Insights (AI/ML)
Reports to: Data Platform & Analytics Manager

About the Role

Irth is building a governed, multi-cloud Lakehouse on Databricks to unlock cross-product insights, enforce data residency, and accelerate AI/ML innovation for our customers.

We are looking for an MLOps/LLMOps Engineer to translate this foundation into scalable, automated, secure, and observable machine learning and LLM services.

You will work closely with Data Science, Data Engineering, Platform, Product, and domain teams to productionize ML and GenAI capabilities supporting Irth’s key industries:

Damage Prevention
Asset Integrity
Land Management
Stakeholder Engagement

This is a pivotal role in establishing reusable engineering patterns for data contracts, lineage, data quality, security, CI/CD, model deployment, monitoring, and operational reliability.

You will help ensure that models and LLM applications move efficiently from experimentation into production—and remain reliable, observable, secure, and cost-effective throughout their lifecycle.

Key Responsibilities

1. Build the ML/LLM Platform on the Lakehouse

Operationalize the complete ML lifecycle—including training, evaluation, packaging, deployment, and monitoring—on Databricks.
Implement ML workflows using the Bronze → Silver → Gold medallion architecture with Delta Lake as the underlying storage layer.
Establish implementation patterns for Unity Catalog model management, preparing model assets for catalog-based governance, lineage, discovery, and access control.
Develop reusable templates for ML/LLM jobs, workflows, and deployment processes.
Create and maintain cluster policies for ML/LLM workloads aligned with enterprise platform guardrails.
Apply platform standards such as:

Private networking
Mandatory resource tagging
Long-Term Support (LTS) Databricks Runtime versions
Secure secrets management
Appropriate compute policies

Establish reusable patterns that allow Data Scientists and ML Engineers to deploy models consistently and safely.

2. Productionize ML & LLM Features

Partner with Data Science and Product teams to productionize models supporting use cases such as:

Excavation and infrastructure risk scoring
Anomaly detection
Predictive maintenance
Geospatial enrichment
Named Entity Recognition (NER) over parcels and easements
Stakeholder communication summarization
Retrieval-Augmented Generation (RAG)
AI-powered assistants and decision-support applications

Design, build, and maintain production-grade LLM and RAG pipelines.
Implement vector search and retrieval architectures using technologies such as Databricks Vector Search.
Deploy and manage model-serving and inference endpoints.
Optimize inference workloads for performance, scalability, reliability, and cost.
Apply optimization techniques such as:

Quantization
Distillation
Prompt and response caching
Retrieval optimization
Batching

Implement batch, streaming, and online inference patterns based on business and latency requirements.
Establish clear service-level expectations and operational SLAs for Priority A/B/C workloads.

3. Engineer Reliability, Security & Compliance into the ML Lifecycle

Integrate data contracts and quality gates into ML and LLM pipelines.
Implement automated validation for:

Schema drift
Null thresholds
Duplicate records
Referential integrity
Data completeness
Feature-quality issues

Implement PII detection, classification, masking, and obfuscation before sensitive data is consumed by features or models.
Enforce data residency requirements through policy-as-code.
Ensure regulated or sensitive Bronze-layer data remains in the required geographic region.
Ensure only appropriately anonymized or aggregated data is transferred to global workspaces or services.
Maintain end-to-end lineage across the ML lifecycle, including:Source tables/columns → Features → Models → Serving Endpoints → Applications/BI
Surface lineage and governance information through appropriate monitoring and governance dashboards.
Support security, compliance, audit, and access-review requirements across the ML platform.

4. Automate Everything – CI/CD & Testing

Use Databricks Asset Bundles (DABs) and GitHub Actions to version, test, and promote ML/LLM assets across environments.
Automate promotion across:DEV → QA → PROD
Ensure production changes are deployed through controlled CI/CD processes with no direct development in PROD.
Version and manage:

Jobs
Notebooks
Model artifacts
Cluster policies
Configuration
Permissions
Deployment definitions

Build automated unit, integration, regression, and data-quality test suites.
Implement model-quality validation as part of deployment pipelines.
Validate business KPIs and analytical outputs against the Unity Catalog semantic layer before production publication.
Establish deployment gates that prevent models or applications from progressing when quality, security, or performance requirements are not met.

5. Observability & Production Operations

Instrument ML/LLM pipelines and services to support defined SLOs, including:

P1 pipeline success: ≥99.5%
P1 MTTD: ≤5 minutes
MTTR: ≤60 minutes

Implement proactive monitoring, alerting, and operational dashboards.
Integrate automated Jira ticket creation for qualifying P1 production failures.
Monitor ML systems for:

Model performance degradation
Data drift
Concept drift
Feature-quality degradation
Prediction distribution changes

Establish LLM-specific observability, including:

Hallucination rates
Response quality
Latency
Token consumption
API usage
Inference costs
Retrieval quality

Define appropriate thresholds and automated alerts for model and LLM quality degradation.
Develop and maintain production runbooks, troubleshooting procedures, and operational documentation.
Participate in or establish appropriate on-call rotations for critical ML/LLM services.
Develop and maintain disaster-recovery procedures aligned with tiered RTO/RPO objectives, including examples such as:

Tier 1: RPO ≤15 minutes / RTO ≤2 hours

Participate in DR testing and document recovery outcomes and remediation actions.

6. FinOps & Cost Management

Enforce mandatory cost and ownership tags across ML/LLM infrastructure, including:

Domain
Tenant
Environment
Cost center
Owner

Ensure tags are consistently propagated across jobs, clusters, warehouses, and other applicable resources.
Support showback and chargeback reporting for ML/LLM workloads.
Monitor compute, storage, model-serving, and LLM/API costs.
Identify opportunities to optimize infrastructure and inference costs without compromising service quality or SLOs.
Establish cost visibility and budget controls for production AI workloads.
Detect and investigate abnormal cost increases or inefficient workloads.

Role Outcomes

In this role, you will help establish the engineering foundation that allows Irth to move from ML/LLM experimentation to reliable production AI at scale.

Success means that:

Models and LLM applications can be deployed through repeatable, automated CI/CD processes.
ML/LLM workloads are secure, governed, observable, and production-ready.
Data, feature, model, and serving lineage is traceable end-to-end.
Production services consistently meet defined reliability and performance SLOs.
Model and LLM quality degradation is detected before it materially impacts customers.
AI workloads are optimized for both business value and infrastructure/API cost.
Data Scientists and Product teams can leverage standardized MLOps/LLMOps patterns rather than building bespoke deployment and monitoring solutions.

Requirements

Qualifications

Required Qualifications

3–5 years of experience in MLOps, LLMOps, ML Engineering, Data Engineering, or platform-focused ML engineering.
Hands-on experience with Databricks, including:

Databricks Jobs and Workflows
Delta Lake
Unity Catalog
Databricks SQL Warehouses

Proven experience building and maintaining CI/CD pipelines for data and ML workloads using:

GitHub Actions
Databricks Asset Bundles (DABs)
Environment promotion across DEV → QA → PROD
Parameterized deployments
Secure secrets management using Azure Key Vault (AKV), AWS KMS/Secrets Manager, or equivalent technologies

Strong understanding of data contracts, schema governance, and automated data/feature validation.
Experience implementing Great Expectations-style validation frameworks or equivalent rule-based data-quality solutions.
Experience building observable production pipelines, including metrics, dashboards, alerting, and monitoring against defined SLOs such as:

Pipeline success rate
Data freshness
Mean Time to Detect (MTTD)
Mean Time to Repair (MTTR)

Strong security-first mindset with practical experience in:

RBAC/ABAC
Unity Catalog security
PII detection and obfuscation
Private networking
Data-access controls
Policy-as-code for data residency

Strong proficiency in Python and SQL.
Working knowledge of distributed computing and job orchestration within Databricks/Spark environments.
Ability to troubleshoot production ML/data workloads and participate in operational support and incident resolution.

Preferred Qualifications

Hands-on experience with LLM/GenAI workflows, including:

Prompt engineering
Retrieval-Augmented Generation (RAG)
LLM evaluation frameworks and evaluation harnesses
AI safety and guardrails
Retrieval and response-quality evaluation
Latency optimization
Token and API-cost optimization

Experience with geospatial data and analytics, including technologies and concepts such as:

PostGIS
Spatial joins
Spatial indexing and tiling
Coordinate systems and projections
GIS-based feature engineering

Experience integrating Power BI with Databricks SQL Warehouses and semantic layers, including an understanding of:

Dataset refresh SLAs
Query concurrency
Row-Level Security (RLS)
Object-Level Security (OLS)

Practical knowledge of FinOps, including:

Resource tagging
Budget management
Cost monitoring
Showback/chargeback
Cost anomaly detection and alerting

Knowledge of Databricks disaster-recovery patterns, including:

Delta Lake Deep Clone
Delta Sharing
Cross-region recovery
Tiered RTO/RPO strategies
DR testing and evidence collection

Hands-on experience with Microsoft Azure and AWS, particularly where ML and data workloads span both environments.
Understanding of cloud-native security patterns, including:

Private Link
VPC/VNet connectivity and peering
Egress restrictions
KMS
AWS Secrets Manager
Azure Key Vault
Data-plane isolation

Ability to work effectively across cloud, platform, data, ML, security, and product teams.

Nice-to-Have Qualifications

Experience deploying and operating models supporting excavation risk scoring, asset integrity, anomaly detection, or predictive maintenance.
Experience building CI/CD workflows that promote asset-integrity or infrastructure-risk models across DEV → QA → PROD, with automated data contracts and quality gates.
Experience implementing model and data observability for workloads using pipeline inspection, sensor, maintenance, or asset-condition data streams.
Familiarity with monitoring:

Data and concept drift
Model performance
SLOs
Pipeline health
Alerting and incident management

Understanding of data residency, security, privacy, compliance, and disaster-recovery requirements for asset-integrity, pipeline, utility, or infrastructure data used by ML services.

Success Metrics

Success in this role will be measured by the engineer’s ability to establish reliable, repeatable, and secure MLOps/LLMOps practices across the enterprise platform.

Key measures include:

Reliable promotion of ML/LLM workloads through DEV → QA → PROD using automated CI/CD.
Consistent implementation of data contracts, validation rules, security controls, and governance requirements.
Production pipelines meeting defined availability, freshness, MTTD, and MTTR SLOs.
Strong observability across data, features, models, LLM applications, and serving infrastructure.
Reduced production incidents through proactive monitoring, automated testing, and standardized deployment patterns.
Effective management and optimization of ML/LLM infrastructure and inference costs.
Demonstrated compliance with security, residency, lineage, and DR requirements.
Reusable MLOps/LLMOps patterns that enable Data Science and Product teams to deploy new AI capabilities faster and more safely.
Strong collaboration with Data Science, Data Engineering, Architecture, Security, Product, and domain teams.

Benefits

What We Offer

Be part of a dynamic and growing company that is well-respected in its industry.
Competitive compensation based on experience and qualifications.
Health Insurance coverage