Integration and Data Architect

Intelligence & Data Architect – Contract – London 5 days per week on site

£900 per day via umbrella

5 days per week on site in central London

About the Project

Building a multi-service platform for financial analysis, portfolio construction, and AI-assisted investment advisory. It pairs a Snowflake-backed data lake with a Neo4j dual-graph (domain ontology + lexical / GraphRAG) so portfolios, instruments, and unstructured research can be queried through one financial DSL. On top of that data substrate sits R.ai, a governed LLM advisor with capability boundaries, approval gates, disclosure, and full audit chains, exposed via streaming chat and voice. The frontend (React/TypeScript monorepo) and backend (FastAPI services on EKS, deployed via Helm and ArgoCD) are wired together with PBAC, OTEL, and a strict ADR/RFC governance process. The platform is in MVP delivery for an enterprise asset-manager client.

Mission Own the end-to-end design from raw data to AI advice: ensure every layer — data assets → ontology / dual-graph → financial DSL → LLM gateway → tool gateway → agent → UX — has a coherent contract, an ADR behind it, and a verification path (formal, runtime, audit). Ship hands-on, slice by slice, while keeping the architecture decisions, ontology, and governance posture defensible to engineering, compliance, and the client.

Required Qualifications

• Building software / data systems with hands-on Python (FastAPI, asyncio, pytest, type hints, monorepo discipli
• ne)Deep data architecture: ontology, taxonomy, conceptual / logical / physical modeling, data contracts, gap analy
• sisKnowledge graph design with Neo4j — dual-graph (domain + lexical), Cypher, sizing (Aura), node/edge versioning, GraphRAG patte
• rnsDomain-specific language design and implementation — grammar, type system, semantic mapping validator, executor, YAML/DSL pipeli
• nesAI agent orchestration — LLM gateway, tool gateway, streaming (SSE + WebSocket), agent ↔ platform contracts, agent S
• DKsResponsible AI governance — capability boundaries, approval gates, disclosure, audit chains to a graph, regulatory T-control traceabil
• ityPBAC + RBAC, JWT auth, security-context propagation, secrets handling, SAST hygi
• eneCloud data platform integration — Snowflake (key-pair auth, schema sync), AWS (EKS, IAM, ALB), Helm, GitOps (Argo
• CD)ADR / RFC authorship and governance — proposed → accepted → superseded lifecycle, registry stewardship, Confluence ↔ Git s
• yncTest discipline — unit + integration + UAT/BDD with enforced coverage thresholds (≥ 90
• %)Strong written communication — ADRs, RFCs, glossary, AGENTS.md, demo playbooks

Desired Experience

• Formal methods — Alloy / TLA+ / model checking for high-assurance compon
• Voice / multimodal AI — STT/TTS via Bedrock or OpenAI, WebSocket pipelines, advisory modality de
• signInformation architecture for documentation — Diataxis, AGENTS.md hierarchy, archival and supersession strate
• giesAI-tooling fluency — Cursor agent skills, MCP servers, prompt engineering, glab/jira CLI automa
• tionCompliance frameworks — regulatory traceability matrices, red-team / adversarial test de
• signObservability — OTEL traces, Phoenix / Grafana, structured logging with rotating hand
• lersModeling languages and ontologies beyond Neo4j (RDF/SHACL, SKOS, financial taxonom
• ies)GitOps / CI quality gates — Bandit, Radon complexity, Angular commit convention, MR review automation

Desired Experience

• Productizing an AI advisor for regulated finance — disclosure → capability refusal → human approval → audit chain, with a compliance-narrator demo to a regulator-style aud
• ienceMigrating a prototype DSL or graph to production — forward-pipeline cutover, mapping validator, deprecation of legacy entry points without breaking labs note
• booksOwning a service from blank repo to client demo — bootstrap, config, health, auth → LLM gateway → tools → governance → voice → Helm chart, all in measurable s
• licesRunning architecture governance for a multi-service platform — 10+ ADRs across data, infra, agents; multiple RFCs; superseding outdated decisions cl
• eanlyBuilding a data-MVP from scratch — scoping data assets (e.g. EODHD, FactSet, Macrobond), mirroring prod schemas in dev, onboarding new vendors via DSL ma
• ppingDesigning human-in-the-loop AI — approval gate, advisory mode, capability boundary, voice with explicit modality co
• nsentAuthoring agent skills / process automation that demonstrably scale a small team's throughput (Jira CLI, GitLab CLI, ADR skills, security-review sk
• ills)Spike-to-decision research — graph store evaluation, GraphRAG vector-store choice, formal verification of high-assurance components, data-locality strategy