Strategy call · The AI Automators · 16 Jul 2026

Where I am — and the question I’m bringing you

A solo build in production for one fund, and how to turn it into modules I can deploy across many — without my overhead scaling with it.

Solo developer 1 live client RAG + n8n in production 30 min

Where I am

I run a full agentic service stack solo. One client is live: Viska Capital, an Icelandic fund. For them I’ve shipped and now operate a production research-and-analysis system — the same class of work you specialise in: n8n production workflows and RAG pipelines.

Everything below is real and running. The reason I’m here is the next step: I built it once, bespoke; I want to build it as reusable parts.

What’s live for Viska

Research engine

Ingestion → RAG

n8n workflows pull sources, embed, and index them into a retrieval store the analysts query.

n8n · RAG pipeline · pgvector*
Chat interface

Mimir

A conversational agent answering over the research corpus — the analyst’s way in.

Mastra agent · Railway · Slack + web
Front end

viska-ai dashboard

The analyst-facing surface: portfolio, themes, positions, and the research read-out.

React / TypeScript · Cloudflare Pages

The system, end to end

CANDIDATE REUSABLE MODULE Sources feeds, docs n8n RAG pipeline ingest · embed · index Research store Viska Research Supabase Mimir — chat agent Mastra · Railway viska-ai dashboard React/TS · CF Pages
Built bespoke for one fund. The dashed span is what I want to lift out as a reusable module.
n8nRAGSupabase Mastra agentsReact / TSRailwayCloudflare Pages

The question I’m bringing you

I’ve built the whole stack once, for one client. How do I turn it into modules I can deploy across more clients — without my overhead scaling with the client count?

Concretely: what to standardise vs keep bespoke, how to package and version the reusable core, and the low-overhead pattern for running many clients as one operator.

What I’d like from you

  1. Modularisation. How do you draw the line between a reusable core and per-client bespoke in n8n production systems — and how do you package and version it?
  2. RAG at scale. Taking one client’s research engine to a repeatable pipeline — what breaks first: ingestion, retrieval quality, eval, or cost?
  3. Low-overhead multi-tenancy. One operator, many clients — shared n8n vs per-client, env isolation, monitoring. What keeps overhead flat?
  4. Solo → scale methods. What development-method shifts (templates, starter apps, CI) do you push for someone at my stage?
  5. Course focus. Given where I am, which parts of the AI Architects Course should I prioritise?
  6. Your read. Candidly — where do I sit in the journey, and what’s my highest-leverage next move?
Dotted items are my inferences from the build, not confirmed — flag any that are wrong: currently just the vector store (pgvector). Everything else is grounded.