Principal AI Data Architect

(AI-Ready Data Platform · ML/LLMOps · Agentic AI Infrastructure · Governance & Security )

Location: Remote

Contract

ABOUT THE ROLE

We are hiring a Principal Data Architect — a hands-on, senior individual contributor who will design, build, govern, and evolve the single source of truth that powers every AI initiative in our organisation. This platform will serve as the foundational nervous system for conversational AI assistants, dashboard intelligence, autonomous AI agents, RAG-powered applications, predictive ML models, and any AI product we build today or in the future.

You will architect the system, drive implementation, own the data contracts that agents and AI applications depend on, enforce security and access governance for both human and agent consumers, and continuously monitor and improve the accuracy and reliability of AI outputs that flow from this platform. You are the person who ensures our AI systems are only as good as the data beneath them — and you make that data exceptional.

WHAT YOU'LL OWN

1. AI-Ready Data Platform — The Single Source of Truth

• Architect and own the enterprise AI data platform — the unified, governed layer that ingests, transforms, stores, and serves all data consumed by AI systems across the organisation.
• Design multi-domain data models (lakehouse, data mesh, event-driven) that are structured from day one to serve AI workloads: clean lineage, versioned schemas, well-documented contracts, and low-latency serving APIs.
• Own the full data stack: real-time streaming (Kafka, Spark Structured Streaming), batch processing (Databricks, PySpark, Delta Lake), cloud storage and compute (AWS, Azure), and data quality / metadata management.
• Ensure this platform is the single, authoritative data source for all downstream consumers — conversational AI, dashboard assistants, autonomous agents, ML models, and reporting — eliminating data silos and conflicting truths.
• Drive modernisation of legacy pipelines (on-prem ETL, batch DWH) to cloud-native, AI-ready architectures with measurable improvements in cost, latency, and delivery velocity.

2. Semantic Models & Knowledge Layer

• Design the semantic layer that sits above raw data — business-aligned ontologies, entity relationships, domain taxonomies, and knowledge graphs — so AI systems understand context, not just tokens.
• Build and maintain knowledge graphs (Neo4j or equivalent) that capture relationships between business entities, policies, KPIs, hierarchies, and domain rules — enabling structured reasoning alongside unstructured retrieval.
• Define and govern a feature store and semantic data contracts that serve both classical ML models and LLM-based applications from a single, well-versioned, trusted source.
• Own metadata management, data lineage, and audit trails across the semantic layer — ensuring every AI system can trace its outputs back to source data with full accountability.

3. RAG, Vector & Retrieval Infrastructure

• Design the retrieval infrastructure that powers RAG-based AI applications: embedding pipelines, vector stores (Pinecone, FAISS, ChromaDB, OpenSearch), chunking strategies, and hybrid retrieval layers combining semantic search with structured queries.
• Define the data contracts between the AI data platform and retrieval consumers — ensuring consistent, freshness-guaranteed, well-indexed data surfaces to RAG pipelines, conversational AI, and agent tools.
• Architect retrieval systems that balance precision, recall, latency, and cost — with clear evaluation benchmarks, not just infrastructure defaults.

4. ML/LLMOps Infrastructure

• Own the ML and LLMOps data infrastructure: training data curation pipelines, feature engineering, model registry, experiment tracking (MLflow), automated evaluation, and production monitoring.
• Build CI/CD pipelines for AI systems: automated data validation, model quality gates, deployment automation, rollback mechanisms, and production health dashboards.
• Design data infrastructure for LLM fine-tuning workflows — training corpus curation, data quality filtering, RLHF pipelines, and adapter management — ensuring models trained on this platform reflect accurate, governed, domain-specific knowledge.
• Establish LLMOps best practices across the organisation: versioning, A/B evaluation, shadow deployments, and canary releases for AI model updates.

5. Multi-Consumer AI Serving Architecture

The platform must reliably serve a diverse set of AI consumers. You will design the serving architecture and data contracts for each:

• Conversational AI Platforms — low-latency, context-rich data APIs that power chatbots, voice assistants, and enterprise copilots with accurate, fresh, source-grounded responses.
• Dashboard Assistants & BI Copilots — semantic query layers and text-to-SQL infrastructure that allow natural language interfaces to query structured business data accurately and safely.
• Autonomous AI Agents — structured tool APIs, function-calling schemas, and memory/state data stores that agents depend on for context retrieval, action execution, and multi-step reasoning.
• Predictive ML Models — feature pipelines, training datasets, and real-time feature serving for classification, forecasting, anomaly detection, and propensity models.
• Ad-hoc AI Experimentation — governed sandbox environments where data scientists and AI engineers can access production-equivalent data safely for research and prototyping.

6. Governance, Security & Access Control

• Design and enforce a comprehensive data governance model that governs access for both human users and AI agents — with role-based access control (RBAC), attribute-based policies, and agent-specific permission scopes that prevent privilege escalation.
• Implement data security controls across the platform: PII detection and masking, data classification, encryption at rest and in transit, audit logging, and compliance alignment (SOX, GDPR, SOC 2, AML/KYC, APAC regulations).
• Define agent data access boundaries — what data an autonomous agent can read, write, modify, or delete — and enforce those boundaries at the platform layer, not just at the application layer.
• Build data contracts and schema governance that prevent upstream changes from silently breaking downstream AI applications, with automated breaking-change detection and versioned migration paths.
• Own regulatory and compliance readiness for all AI data pipelines — ensuring audit trails, explainability artefacts, and data provenance are available on demand.

7. Agentic Behaviour Observability & Output Accuracy

• Own the observability stack for AI agent behaviour: instrument agents to capture inputs, retrieved context, tool calls, reasoning traces, and outputs — creating a complete audit trail of every agentic action driven by platform data.
• Design and operate evaluation frameworks that continuously measure AI output quality: factual accuracy, context faithfulness, retrieval relevance, hallucination rates, and task completion success — across all AI consumers of the platform.
• Establish feedback loops between evaluation signals and platform improvements: when agent outputs degrade, trace the root cause to data freshness, retrieval failures, schema drift, or model issues — and own the remediation.
• Define SLAs for AI output quality and data freshness; build alerting and escalation frameworks that surface platform-driven AI degradation before end users notice.
• Implement human-in-the-loop review workflows for high-stakes agent actions — ensuring critical decisions have appropriate oversight, audit records, and rollback capability.

8. Architecture Standards & Engineering Enablement

• Define and maintain the reference architecture for the AI data platform — documenting design patterns, data contracts, integration standards, and decision records (ADRs) that all engineering teams follow.
• Establish data engineering standards: pipeline testing frameworks, code review practices, CI/CD automation, infrastructure-as-code (Terraform), reusable component libraries, and observability instrumentation.
• Serve as the senior technical reviewer for all data system designs that interact with the AI platform — ensuring consistency, security, and quality across every integration point.
• Run internal architecture workshops, design reviews, and enablement sessions to embed AI-ready data platform best practices across data engineering and AI teams.

Must-Have Experience:

• 15+ years of hands-on data engineering and architecture experience, with 3–5+ years building production AI/ML and LLM-era data infrastructure.
• Proven experience designing enterprise-scale AI data platforms that serve multiple AI consumers — not just one application or pipeline.
• Deep expertise in lakehouse and data mesh architectures: Databricks, Delta Lake, PySpark, Kafka, Spark Structured Streaming, cloud-native data services (AWS, Azure).
• Hands-on experience with vector stores, semantic models, knowledge graphs, and retrieval infrastructure in production environments.
• Working knowledge of LLMOps: model serving pipelines, MLflow, CI/CD for AI, automated evaluation, and production monitoring.
• Strong background in data governance, security, and compliance in regulated industries (financial services, payments, cybersecurity, healthcare).
• Experience defining data access controls for AI agents and automated systems — not just human users.

Technical Skills:

• Expert: Python, SQL, PySpark, Kafka, Databricks, Delta Lake, Snowflake, AWS (S3, Glue, EKS, Bedrock, Kinesis, Redshift), Docker, Kubernetes, Terraform, GitHub Actions.
• Strong: LangChain, LlamaIndex, LLM APIs (OpenAI, AWS Bedrock, Claude, HuggingFace), vector databases (Pinecone, FAISS, ChromaDB, OpenSearch), knowledge graphs (Neo4j).
• Solid: MLflow, FastAPI, CI/CD pipelines, observability tooling (CloudWatch, Grafana, or equivalent), data lineage and metadata management platforms.

The Right Mindset

• You think platform-first: every design decision considers all current and future AI consumers, not just the one use case in front of you.
• You are as comfortable in a governance design session as you are debugging a broken embedding pipeline at 11pm.
• You believe agent safety starts with data — and you design access controls, audit trails, and guardrails before the agents are built, not after.
• You close the loop: you do not consider a feature shipped until you have observability, evaluation, and a feedback mechanism in place.

PREFERRED QUALIFICATIONS

• B. Tech / M. Tech / M.S. in Computer Science, Information Technology, Data Engineering, or a related quantitative field.
• Experience in presales solutioning for large data and AI programmes: RFP/RFI responses, SOW shaping, effort estimation, CXO-level solutioning.
• Prior experience at a global financial institution (payments, risk, AML, compliance) or enterprise SaaS — where AI data infrastructure operated under strict regulatory oversight.
• Familiarity with emerging agent infrastructure standards: MCP (Model Context Protocol), agent memory architectures, multi-agent coordination frameworks (LangGraph, AutoGen, CrewAI).
• Experience designing observability and evaluation systems specifically for agentic AI behaviour, not just traditional software or ML model monitoring.