If the only way to get an answer in your Amazon business is to ask the founder, that is a bottleneck RAG (Retrieval Augmented Generation) is built to remove. Shubhash explains the five-step RAG pipeline — chunk, embed, index, retrieve, answer — built on tools most sellers already have (Postgres/Supabase), while Danny covers the model-switching setup he built to cut routine AI costs by roughly 90% without disrupting his main workflow.
Key Topics
- RAG systems for Amazon sellers – chunking, embedding, indexing, retrieval and answer, built on Supabase/Postgres with pgvector
- Data drift and guardrails – how to stop a RAG system guessing, and how to version-control its answers over time
- Model switching for cost control – why and how, via OpenClaude, OpenRouter and DeepSeek
- Harnesses vs models – why a model performs differently outside its native environment
- OpenMontage – an AI video editing repo, and why domain experience still can’t be replaced
Timestamps
- 00:00 – Danny opens week three, hands over to Shubhash
- 00:45 – Shubhash introduces today’s topic: RAG (Retrieval Augmented Generation)
- 01:39 – Why sellers should build a RAG: founder-bottlenecked knowledge
- 02:53 – No model training, no GPUs required – “organising a warehouse, not building a robot”
- 03:42 – The five steps: chunk, embed, index, retrieve, answer
- 05:01 – Indexing on Postgres/Supabase, searchable by meaning and by exact keyword
- 06:08 – Retrieval and the router: number questions to tables, everything else answered with sources
- 06:54 – Danny on data drift: the chicken-and-egg problem of trusting Claude to catch its own drift
- 08:18 – Core of it simplified: chunk, guardrails, plain-English queries via Slack or Claude
- 09:15 – Building on the same Supabase project from earlier sessions, switching on pgvector
- 10:12 – Guardrail rule: don’t guess, cite sources, flag when there’s no answer
- 10:41 – Version control: feeding user feedback back into the system
- 11:27 – Danny’s segment: model switching, why and how
- 13:22 – The CCR experiment: a gateway/pass-through that didn’t preserve context or dependencies
- 15:14 – Why OpenClaude was the better route: terminal-based, dependencies not preloaded
- 16:44 – DeepSeek vs Fable 5: close, not equal, but roughly 90% cheaper for grunt work
- 17:40 – Harnesses matter as much as the model itself
- 20:38 – Budget-capping your OpenRouter API key (and the horror stories of not doing so)
- 22:52 – Why Danny moved off Claude in Chrome for email rebuilds – context window visibility problems
- 23:37 – Solving it with dual Google Workspace access via CLI (Seller Sessions + DataBrill)
- 25:17 – OpenMontage: plug in the repo, but expect to still do heavy lifting
- 27:36 – The aggregator analogy: why “done for you” video/design without domain experience is a scam
- 30:46 – Round-up: three steps to start a RAG system yourself
- 34:56 – Where to reach Shubhash
Key Takeaways
- RAG breaks the founder bottleneck – your spreadsheets and documents become queryable with sources attached, cutting out the “ask the founder” loop.
- RAG is simpler than it sounds – five steps (chunk, embed, index, retrieve, answer), no model training, no GPUs, built on Postgres/Supabase you likely already have.
- Guardrails beat guessing – if there’s no answer in the data, say so rather than fabricate one.
- Model switching only pays off with the right setup – a terminal-based OpenClaude + OpenRouter + DeepSeek combo saved roughly 90% on grunt-work token spend.
- AI video editing still needs domain experience – OpenMontage is a genuine leap forward, but a done-for-you button-press is the same mistake the Amazon aggregators made.
Notable Quotes
“The business knowledge lives in the founder’s head and a stack of spreadsheets that only they can navigate.” – Shubhash
“It’s easier or better to say, I don’t have enough data to make a conclusion to this, than to guess.” – Shubhash
“Harnesses are just as important, to a point, as the model itself.” – Danny McMillan
“If you said to a video editor, I’ll just press a button and it makes me a video – they’d say that sounds like a scam to me.” – Danny McMillan
Resources Mentioned
- Supabase / Postgres + pgvector – the database layer for storing and searching RAG “cards” by meaning and keyword
- OpenRouter – model-switching gateway used to route to DeepSeek; supports API key budget caps
- DeepSeek (v4 Pro) – the model Danny settled on for cost-saving grunt work, roughly 90% cheaper than Fable 5
- OpenClaude – terminal-based alternative to a full CCR gateway, dependencies loaded on demand
- Google Workspace CLI – used to run two email accounts through Claude without a browser context-window problem
- OpenMontage – GitHub repo for AI-assisted video editing, works alongside a Claude video/Remotion flow
Connect
Shubhash – Not a Square
Email: shubhash [at] notasquare.io
Seller Sessions is the leading podcast for advanced Amazon sellers, hosted by Danny McMillan. Week four returns with Ritu for Go With The Flow.