AI Solutions Architect — Company Memory & Executive Intelligence Layer (Claude, MCP, RAG, n8n)

Summary We are looking for a hands-on AI Solutions Architect and systems builder to design and build an internal AI operating layer for a growing creative, production and marketing services group. This is not a generic chatbot project, not a prompt-engineering exercise, and not simply a finance dashboard, SOW tracker or workflow-automation job. Automation is one tool we will use, not the point of the work. The ambition is a company memory and executive intelligence system: a structured, searchable institutional memory across the business that lets senior management retrieve context, connect information across systems, spot patterns, preserve knowledge and make faster, better-informed decisions. Alongside it, we want a clear view of where AI, agents and automation can make each core function more effective. What we want to build There are two connected objectives. First, the company memory and executive intelligence layer: a shared institutional brain that lets one or two senior people interrogate the whole business across documents, systems and workflows, and lets functional leaders retrieve the history and context relevant to their area. Second, a prioritised map of where AI agents, retrieval, document generation, reporting and task-specific copilots can make each discipline faster, more consistent and more effective. We are not looking for one mega-assistant. We want a thin layer to ask across the business, with a small number of function-scoped agents beneath it, each grounded in the right source material and each only able to see what its role should see. Our environment Knowing our actual stack up front will make your proposal far stronger: Roughly 90 percent of our knowledge sits in Google Workspace (Drive, Gmail, Docs, Sheets, Calendar). This is the centre of gravity. Slack for internal chat, not documents. Accounting on Xero, across seven separate company entities. Project and resourcing data in Streamtime, currently hand-entered and not connected to accounting, so figures are re-keyed by hand. Some email on Outlook and a legacy Microsoft setup, mostly in finance. AlphaSense and external data sources for new-business and market research. We are exploring Claude Desktop, Claude Code, MCP connectors, n8n, API-based workflows and RAG / document retrieval. Context and constraints you should already be thinking about The proposals that stand out will engage with the hard parts, not just the happy path: Data quality. Some sources (Streamtime in particular) are hand-entered and will not all be reliable enough to build on. Deciding what to trust is part of the work. Source of truth. When Drive and Xero disagree, the system has to know which one wins. We need a defined source hierarchy. Access and security. Board, investor and M&A material is sensitive. We need role-based access from day one, so the right people can ask anything while others see only their slice, and our data stays in our control. Seven entities. Pulling Xero together across seven companies, with the right figures attributed to the right entity, takes real work to get right. Maintainability. We want an approach we can understand, maintain and extend internally, built on documented, non-proprietary foundations wherever possible. We are wary of being locked into bespoke infrastructure only one person can run. Functional areas in scope Finance: reporting, cashflow, debtor and margin analysis, forecasting, board and investor reporting. Legal and commercial: SOW review, contract retrieval, scope comparison, change orders, approval language. Client service and engagement: client history, meeting preparation, status reporting, action tracking, relationship intelligence. Production and delivery: timelines, suppliers, budgets, risk tracking, deliverables, dependencies, post-project learnings. New business: prospect research, proposal and credentials generation, pitch support, follow-up. Strategy and creative: category intelligence, brand and prior-work retrieval, deck support, idea development. HR and resourcing: availability, role requirements, onboarding, policy retrieval, continuity planning. Leadership and operations: weekly business intelligence, management reporting, escalation tracking, decision support. Questions the system should help answer What did we agree with a given client, partner or investor, and when? What was decided in previous board discussions on a particular opportunity? Which clients, projects or opportunities show the strongest margin, growth or repeat potential? Which SOWs, project types or clients have historically created budget, scope or delivery problems? For example, where a SOW says £100k and Xero shows £90k billed, surface the £10k gap ready for someone to act on. What prior work, contracts or client history should inform a new decision or a new brief? What context would be lost if a key person were unavailable for several weeks? What should senior management know this week that is not obvious from any single document or meeting? Which workflows are too manual today, and where can AI reduce repetitive chasing, reporting, document assembly and retrieval? Who we are looking for Someone who thinks architecturally and builds practically. You can assess the current landscape, recommend the right approach, connect systems, build and test outputs, document clearly, and hand over something genuinely usable by senior, non-technical people. We are not looking for: generic AI consultants; prompt engineers without systems-building experience; website-support chatbot developers; machine-learning researchers or model trainers; pure DevOps engineers; strategists who cannot build; automation freelancers who treat this as only an n8n or dashboard job; or anyone who thinks about one department rather than the operating architecture of the whole business. Required skills Claude Desktop and Claude API, ideally Claude Code MCP (Model Context Protocol) AI agents and tool-using AI workflows RAG, document retrieval and internal knowledge-base design Source-hierarchy and data-architecture design (deciding what is authoritative) Google Workspace APIs: Drive, Gmail, Calendar, Docs, Sheets Data extraction from documents, spreadsheets, PDFs and structured files REST APIs, OAuth, webhooks and API authentication n8n workflow automation Slack API or Slack workflow automation Strong documentation and handover discipline Useful but not essential: Microsoft Graph / Microsoft 365 (Outlook, SharePoint, OneDrive), since our Microsoft footprint is mostly legacy finance; Xero API experience; vector databases such as Supabase / pgvector, Pinecone, Weaviate or Chroma. Programming Python TypeScript / JavaScript / Node.js SQL Google Apps Script preferred Comfortable with JSON, YAML, Git, command line tools and API testing Nice to have Built internal AI copilots, company-memory systems, executive intelligence tools or reporting agents Worked with agency, consulting, creative, production or professional-services businesses Finance, project, board or executive reporting automation Designed internal systems for senior management users Docker, LangChain or LlamaIndex, Retool, Airtable, Make or Zapier The first phase We want to start with a paid discovery, architecture and prototype-planning phase, likely 2 to 4 weeks, with a clear go/no-go before any larger build. The first phase should produce: A review of our systems, tools, folders, data sources and knowledge assets, with an honest read on data quality. A review of key functional workflows across the business and where AI or automation creates the most leverage. A recommended architecture for the company memory and executive intelligence layer. A clear split of what should sit in Claude, MCP, n8n, custom code, APIs and our existing tools. A defined source hierarchy and access model, including how the system knows which documents and data are authoritative. A prioritised list of candidate first use cases, and a recommendation on which to prototype first and why. A proposed build plan, timeline and technical approach. A prototype specification: inputs, outputs, workflows, success criteria and limitations. A roadmap for extending the system into function-specific AI workflows or copilots. Documentation that both technical and senior business readers can follow. We want someone who will challenge the brief and recommend the best path, not simply build what we describe. Please do not assume the first prototype should be finance, SOWs or project reporting. Tell us how you would decide. How to apply Please include: A short summary of your relevant experience. Two or three examples of comparable AI automation, company-memory, RAG, agentic-workflow or business-systems projects you have built. Your experience with Claude, Claude Code, MCP and n8n. Your experience connecting Google Workspace and Slack, and Microsoft 365 if relevant. The languages and frameworks you would use, and why. How you would approach the first 30 days, and how you would assess our systems, data and workflows before recommending a first build. The criteria you would use to prioritise the first prototype. The architecture you would recommend at a high level. The main implementation challenges or failure points you would expect. Whether you are stronger as a builder, architect or strategist, plus your availability and preferred working model. We are UK-based, so strong English and at least 3 to 4 hours of daily overlap with UK working hours is required. We are looking for a practical builder who can move quickly, think commercially, communicate clearly, and create something genuinely useful to senior management and key functional teams.