Seller Sessions Amazon FBA and Private Label

Description: Danny McMillan and Ritu Java on AI-built email triage, pest control and client sentiment analysis for busy Amazon agencies and sellers.
Ritu Java returns after recovering from a serious ankle injury, and dives straight into the theme of the episode: managing cognitive load as AI accelerates faster than we can adjust to it. You'll hear how Danny and Ritu are each solving the same problem from different angles — building systems that manage the noise so the signal gets through.
Ritu walks through two Claude-built systems running on her Google ecosystem: a "pest control" routine that reads her inbox twice daily and files unsolicited pitches before she ever sees them, and a client sentiment analysis system that reads years of email history to flag which client relationships are improving or deteriorating. Danny shares his own zero-inbox protocol — a labelled morning triage that clears by 8:33am UK time — and both agree the real skill now is boundaries, not more automation.
Key Topics
AI acceleration and cognitive load - why keeping pace with AI now outstrips our own working speed
Pest control inbox routine - a Claude + Gemini API system that filters unsolicited email using "left brain, right brain" logic
Reply radar - surfacing the emails that genuinely need a response, out of tens of thousands
Zero inbox protocol - Danny's labelled, time-boxed morning triage system
Client sentiment analysis - mining historical email threads to score relationship health
Cascading plans - reviewing AI-generated plans in stages instead of all at once, to protect decision quality
Timestamps
[00:00] Ritu returns after an ankle injury, catches up on the last few months
[01:34] The acceleration of acceleration - falling behind AI's own pace
[02:29] Danny's all-in-Claude mandate for his team
[05:36] Cognitive load and the need for "a skill to manage the other skills"
[06:39] Danny on verification systems, sign-off protocol and decision fatigue
[07:19] Cascading plans: reviewing complex work in stages, not all at once
[08:05] Why a shorter, higher-quality working day beats a fixed nine-to-five
[10:03] Ritu's two access routes into Google: MCP connector vs CLI
[10:56] Claude routines explained - the "cron job" for your inbox
[11:55] The pest control system: left brain keyword matching, right brain Gemini analysis
[16:04] Reply radar - catching the emails that were missed
[18:07] Danny's zero inbox protocol: labels, triage timing, draft handling
[20:39] The "sixth follow-up email" rant - boundaries with cold outreach
[23:13] Why fuzzy logic beats simple Gmail filters
[24:17] The CLI route: BigQuery access for agency-wide ad and rank data
[24:59] Client sentiment analysis: reading 500+ emails to score relationship trust
[27:17] Danny on spotting relationship drift before the client raises it
[30:12] Wrap-up: signal to noise as the theme of the episode
Key Takeaways
AI is accelerating faster than we can adapt - the gap between AI's pace and our own decision-making speed is the new operating reality.
Fuzzy logic beats static rules - Ritu's pest control system combines keyword detection with AI judgement, catching what a plain Gmail filter would miss.
A time-boxed triage beats an always-on inbox - Danny's zero inbox protocol proves structure, not more tools, is what protects focus.
Sentiment analysis surfaces relationship drift early - mining historical email tone gives an objective signal alongside human account management.
Cascading plans protect decision quality - reviewing AI output in stages, not all at once, avoids the fatigue that degrades judgement.
Notable Quotes
"We're slower than the AI now in terms of how fast things are moving." - Ritu Java "AI is not an enabler for us to think on our behalf. It's our job to verify it." - Danny McMillan "In order for us to be efficient with our productivity, we need something to manage us." - Ritu Java "Without perspective and perception, empathy can't live." - Danny McMillan
Resources Mentioned
Claude routines - scheduled, cron-style Claude tasks used to run the pest control and reply radar systems
Gemini API - powers the "right brain" analysis layer in Ritu's pest control system
Google Workspace MCP connector - built-in Claude access to Gmail and Drive
CLASP - CLI tool for Google Apps Script development
BigQuery CLI - command-line access to agency ad, organic and rank-tracking data
Connect
Ritu Java - AI for E-commerce newsletter; speaking at the Amazon booth this month
Seller Sessions is the leading podcast for advanced Amazon sellers, hosted by Danny McMillan. Adam Heist returns next week for the Broadmatch Show.

Week three of the all-things-Claude series brings Shubhash back after some time away. He walks through why sellers should build a RAG (Retrieval Augmented Generation) system: business knowledge trapped in a founder's head and a stack of spreadsheets creates a bottleneck, and RAG breaks it by answering questions from your own files with sources attached, no hallucination, no SQL required.
Danny follows with a practical rundown of the model-switching setup he's spent the last week building: a Claude Code Router (CCR) experiment that didn't work out, and the eventual fix via OpenClaude, OpenRouter and DeepSeek to save on token spend for grunt work without disturbing his main Claude Desktop workflow. He also covers running a second email address on a different domain via the Google Workspace CLI, and gives an honest verdict on OpenMontage for AI-assisted video editing: impressive, but still not a replacement for editing domain experience.
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.

Danny McMillan returns after his longest break in almost ten years, with Seller Sessions approaching its tenth anniversary and roughly 1,300 episodes. This is the pilot of a new monthly roundtable with Sim and Matt (Dorian returns next month), moving away from the conversion show format towards raw conversation.
You'll hear how Sim's team runs product development end to end with AI: keyword-scored idea validation, brand director sign-off, Claude-generated product concepts rendered through Codex, and a launch pipeline already booked out to 2027. Matt shares how Productpinion prioritises features from customer feedback, and why prioritisation is the most undervalued skill in the AI era.
The back half tackles the big theme: cognitive load. Danny breaks down verification fatigue, context switching and AI maxing, and why the scarce resource is no longer time but attention and decision quality.
Key Topics
AI-driven product development - from keyword scoring to Claude SVG concepts and Codex-generated product renders
Team structure at scale - how ideas route through brand directors to sourcing across UK and Philippines teams
Hiring in the AI era - why refusing to use AI is now a dealbreaker, and why gutting teams for AI is commercial suicide
Free local AI tools - Fluid Voice (Whisper Flow alternative) and Meetily (Granola alternative)
Cognitive load and verification fatigue - the hidden tax of moving from doer to overseer
Timestamps
00:00 - Danny returns: ten years of Seller Sessions, new pilot format
01:50 - Sim's update: ditching ClickUp for a bespoke operating system
03:55 - Matt's update: closing the research loop in Productpinion, Florence CRO brain, MCP
05:39 - Sim's product pipeline: AI keyword scoring, brand director approval, deep research
07:07 - AI product development: Claude concepts, Codex renders, 3-in-1 product mashups
09:07 - Packaging designed for the main image, and how far you can push it
10:46 - Team workflow: brand directors owning P&L, sourcing handoffs
14:21 - Danny on gutting teams for AI: who maintains the machines?
15:56 - The hiring line: refuse to use AI, you haven't got a job
17:48 - Claude across every department: projects, Claude Code vs Cowork
20:10 - Free local tools: Fluid Voice for dictation, Meetily for meeting notes
21:57 - Marketplace arbitrage: moving proven products between Amazon marketplaces
23:21 - Matt on signal to noise: wasting tokens instead of wasting time
24:57 - Time blocking and prioritising features by customer impact
26:54 - Danny's segment: cognitive load, oversight duty and verification fatigue
31:46 - Asking the right question: the Claude Science deep-dive example
36:41 - AI maxing, context switching and high-stakes decision quality
42:07 - Claude telling you to go to sleep
45:16 - Danny's framework: reject the first plan, decision sprints, deliberate decompression
50:47 - Where to reach Sim and Matt
Key Takeaways
AI has made product development fun again - unique product concepts generated with Claude and Codex, feeding a pipeline mapped to 2027.
Augment, don't replace - if someone was worth hiring, AI should multiply their output, not justify cutting them.
Prioritisation is the undervalued AI skill - just because you can do everything doesn't mean you should.
Verification fatigue is real - build in decision sprints and deliberate decompression, and reject Claude's first plan on sight.
The scarce resource is attention, not time - your night schedule and recovery feed the next day's output.
Notable Quotes
"AI is enabling the boring to get released and the fun stuff to happen." - Sim
"Instead of people wasting time, now they're just wasting tokens." - Matt
"Prompts don't matter, but asking the right question unlocks everything." - Danny McMillan
"AI doesn't just speed you up. It puts you on permanent oversight duty, and the cost of that duty is your attention and your judgment, not time." - Danny McMillan
Resources Mentioned
Fluid Voice - free, open source local dictation with on-device models; a Whisper Flow alternative
Meetily - free, open source meeting summariser that runs privately on your machine; a Granola alternative
Claude / Claude Code - the AI platform used across both Danny's and Sim's teams
Codex - used alongside Claude to generate product concept images
Productpinion - Matt's shopper testing platform, now with MCP support and draft polls
Connect
Sim - on LinkedIn (genuine reach-outs answered)
Matt - on LinkedIn or via productpinion.com
Seller Sessions is the leading podcast for advanced Amazon sellers, hosted by Danny McMillan. Dorian returns next month.

Building Repeatables in Claude: Skills, CLI vs MCP and Token Discipline | Go With The Flow
Claude Skills, CLI vs MCP and Token Discipline with Ritu Java | Seller Sessions
SEO Description
Ritu Java and Danny McMillan on building agentic skills, choosing CLI over MCP, plan mode discipline and the short window to ship before token costs reset.
Episode Summary
Week 4 of the month, Go With The Flow, and Ritu Java is back from her travels. The world has shipped fast since the last episode: Codex 5.5, Claude 4.7, an Amazon Ads MCP and a fresh round of panic over the rumoured removal of Claude Code from the $20 plan (it was a 2% AB test, not a rollout). Ritu and Danny use the noise to make a sharper point: this is the moment to stop chasing models and start building repeatable systems on the platform you have already chosen.
Ritu walks through the three eras of PPC Ninja's automation stack. Apps Script bulk file generators three years ago, Netlify hosted UI apps last year, and now agentic skills that her team chats with in plain English to produce upload ready Amazon bulk files. The same shift applies to data: BigQuery accessed through the Google Cloud CLI rather than through MCP, because CLI is leaner on tokens and works better when the job is heavy on data rather than tool surface. Danny mirrors the move with his event-ops CLI for WordPress, WooCommerce, Stripe and FooEvents reconciliation, and his four tier ExtractFlow cascade (HTTP, headless, stealth, agentic) that bypasses the limits of any single browser tool.
The second half is a discipline talk. Plan mode every time. Push back on the first plan because Claude over engineers by default. 30% of your time on workflow scaffolding so the other 70% can be real building. The 21 day Claude rule: when a shiny new tool fires the dopamine, wait 21 days before refactoring around it. Left brain tasks (counting, SQL, deterministic logic) belong in scripts. Right brain tasks (judgment, creativity, hypotheses) belong in the model. Mix them inside a single skill. Skills are micro pieces of your workflow, not magic, and Claude can write them for you from an existing SOP.
Key Topics
The three eras of PPC Ninja automation: Apps Script, Netlify UI apps, agentic skills
CLI vs MCP: when to choose each and why CLI is more token efficient for data heavy work
Token economics, the rumoured $20 plan change and why it was a 2% AB test
The short window before subsidised tokens get repriced
Plan mode discipline and the "push back on plan one" rule
Danny's 30 / 70 framework: workflow scaffolding vs building
The 21 day Claude rule for resisting tool churn
Left brain vs right brain task design inside a single skill
The PPC Ninja "5 Whys" skill: deterministic SQL plus non deterministic hypotheses
Claude.md, Gemini.md, Skills.yaml and the emerging Agents.md standard
Skills for beginners: let Claude write them from your SOP
Skill cascading: research, article, LinkedIn post, tweets, slide deck in one chain
Timestamps
[00:01] Welcome back, Week 4 Go With The Flow, Ritu returns from travels
[00:17] Codex 5.5, Claude 4.7 and the "no one is writing code anymore" reality
[02:01] Ritu on the three eras of PPC Ninja automation
[02:42] Era 1: Apps Script bulk file generators in Google Sheets
[03:46] Era 2: Netlify hosted UI apps with input fields
[04:48] Era 3: Agentic skills, the bulk file skill trained on Amazon templates
[06:22] Cloud talking to BigQuery through the Google Cloud CLI
[07:00] Danny: what is a CLI and why it matters for token use
[08:00] Amazon Advertising MCP vs CLI based access to the same data
[09:33] WordPress horrible to drive via MCP, easy via CLI
[10:00] Danny's event-ops CLI: tickets, food tickets, WooCommerce, Stripe reconciliation
[12:13] ExtractFlow four tier cascade: soft, medium, stealth, agentic
[13:46] Why CLI for the heavy stuff, MCP for the soft touch
[14:13] AWS CLI: chat to Claude, push HTML blog posts live in two minutes
[15:33] The overwhelm problem and the 5,000costbehindthe5,000costbehindthe100 plan
[17:35] The $20 plan rumour: it was a 2% AB test, not a rollout
[19:38] Build repeatables, not one offs
[20:38] Danny: pick a platform and stop chasing benchmarks
[21:16] The 21 day Claude rule for new tools
[22:16] Plan mode every time, push back on plan one, get the second plan
[23:02] Why am I building it, who is it for, what am I building
[23:30] The 30 / 70 split: workflow scaffolding vs real building
[25:13] Why long six to fourteen hour Claude runs are usually inefficiency
[27:12] Compounding 1% a day across a year
[27:47] "I build the things that build things"
[28:00] Architecture vs apps: filling the gaps between A and B
[29:06] Left brain vs right brain task design
[30:01] Why throwing 80/20 at a sales drop diagnosis fails
[31:33] The PPC Ninja 5 Whys skill: deterministic plus non deterministic in one flow
[34:32] Claude.md, Gemini.md, skills.yaml and the agents.md standard
[40:53] Beginners: let Claude write the skill from your SOP, use the interview pattern
[42:39] Skill cascading: URL to research to article to LinkedIn post to tweets to slides
[44:42] Mixing deterministic and non deterministic inside a single skill
[45:39] Wrap up, signal to noise, who is it for
Key Takeaways
Pick a platform and stop chasing models. A new model ships every week. Time spent benchmarking is time not building. Double down on Claude (or whichever you chose), use the 21 day rule, and let the ecosystem catch up to the shiny thing in your feed.
CLI for heavy work, MCP for soft touch. MCP loads tools and skills into context and burns tokens. CLI uses programs already on your machine. For data heavy jobs (BigQuery, AWS, WordPress at scale), CLI wins. For light cross app workflows, MCP is fine.
Build repeatables, not one offs. Subsidised tokens will not last. The 100planreportedlycostsAnthropic100planreportedlycostsAnthropic5,000 to serve. Spend the window building scaffolding that compounds, not 14 hour vibe coding runs.
Plan mode every time, then push back. Claude over engineers by default. Generate the plan, then say "you have over engineered this, although I want it elegant, go back and review." Plan two is the one you start from.
30% on workflow, 70% on building. Each new dependency, MCP, skill or repo you add to your workflow compounds across every future project. Stop building only the apps. Build the things that build the apps.
Left brain in scripts, right brain in the model. Counting, SQL, deterministic logic belongs in Python the moment you can offload it. Save the model for hypotheses, judgment and creativity. The PPC Ninja 5 Whys skill mixes both inside one flow.
Skills are micro pieces, not magic. Take an SOP, ask Claude to interview you with decision panels, and let it write the skill. Then cascade skills together: URL to research to long form article to LinkedIn post to tweets to slide deck.
Notable Quotes
"Instead of doing one offs, it is time to build repeatables. The more people can learn that skill now, the better it will be, because a year from now you may not have access to the same tokens."
Ritu Java
"If you see something and it looks sexy and it has sex and sizzle and your dopamine is screaming to go after it, wait 21 days. Either Claude will have it, or someone will have a repo, and you can combine it."
Danny McMillan
"Always use plan mode. Never accept plan number one. Tell Claude: you have over engineered this, although I want it elegant, go back and review. Then start from plan two."
Danny McMillan
"I build the things that build things. I build the scaffolding the team needs so they can build on top of it."
Danny McMillan
"Spend 30% of your time on your workflow and 70% building. The 30% compounds across every project."
Danny McMillan
"If we just hand six months of ad, organic, ranking and SQP data to Claude with no structure, it is going to mess up. It will give you an 80/20 you are not satisfied with, because it is not equipped to handle that volume without scaffolding."
Ritu Java
"WordPress is horrible to work with through MCP. It falls over all the time. CLI can be amazing for certain things."
Danny McMillan
Resources Mentioned
PPC Ninja : Ritu's Amazon PPC software and agency, base for the BigQuery + CLI stack discussed
Claude Code : Anthropic's CLI for Claude, the primary surface used in the episode
Anthropic Claude : Claude 4.7 referenced as the current model
OpenAI Codex : Codex 5.5 mentioned as the rival shipping fast
Google Gemini CLI : Referenced as a sibling agent surface (Gemini.md)
Google BigQuery : PPC Ninja's central data warehouse
Google Cloud CLI (gcloud) : The CLI Claude uses to talk to BigQuery
Amazon Advertising MCP : Amazon's official MCP server for ads data, referenced as the MCP comparison point
AWS CLI : Used by Ritu to publish HTML blog posts to ppcninja.com from a Claude chat
Netlify : Hosting layer for PPC Ninja's previous era of UI based apps
WordPress and WooCommerce : Backbone of Danny's event-ops CLI
FooEvents : Ticketing plugin that lives behind WooCommerce in the event-ops flow
Stripe : Source of the card fee variation Danny reconciles via CLI
ExtractFlow / CloudExtract : Danny's four tier extraction cascade (HTTP, headless, stealth, agentic). Open repo
Playwright : The default browser automation tier inside ExtractFlow
Agents.md : Emerging AI agnostic instruction file standard alongside Claude.md and Gemini.md
Sequential Thinking MCP : The MCP Danny invokes when asking Claude to step through analysis
Hosts
Danny McMillan : Host of Seller Sessions, founder of DataBrill, building AI native tooling and CLI based workflows for Amazon sellers.
Website: https://sellersessions.com
LinkedIn: https://www.linkedin.com/in/dannymcmillan
Ritu Java : CEO and co founder of PPC Ninja, Amazon PPC software and agency. Specialises in automation, BigQuery pipelines and agentic workflow design.
LinkedIn: https://ca.linkedin.com/in/ritujava
Website: https://www.ppcninja.com
What's Next
Next week: Ritu and Danny pick up routines and the new Claude scheduler.
In 8 days: Seller Sessions Live 2026 in London on 9 May. Last week to lock in any final discounts.
About Seller Sessions
Seller Sessions is the leading podcast for serious Amazon sellers, hosted by Danny McMillan since 2017. Go With The Flow is the weekly automation strand where Danny and Ritu Java work through agentic flows, MCPs, CLIs and skills, in real time, on the same stack their teams ship every week.
Episode published: 1 May 2026
Series: Go With The Flow (Week 4 of the month)
Keywords: claude skills, claude code, cli vs mcp, mcp model context protocol, claude 4.7, codex 5.5, amazon ppc automation, bigquery cli, agentic workflows, plan mode, token optimisation, claude.md, agents.md, ppc ninja, ritu java, seller sessions podcast, go with the flow

Why Your Amazon Dashboard Is Lying to You + Remotion & Voice Cloning Reality Check | Claude Sessions
Amazon Dashboard Brain, Remotion Video & ElevenLabs Voice Cloning | Claude Sessions
SEO Description
Shubhash Sharma on building a data brain behind your Amazon dashboard. Danny McMillan on Remotion video and ElevenLabs voice cloning realities.
Episode Summary
Week 3 of the month means Claude Sessions, and Danny McMillan and Shubhash Sharma are back with a double feature for Amazon and TikTok Shop sellers building their own AI tooling. Shubhash picks up from last episode's SP API and Ads API walkthrough with a hard lesson learned the wrong way: a polished dashboard wired straight into Amazon is a window with no room behind it. The numbers will lie, and you will not know when a feed silently dies.
He walks through the fix: a "brain" sitting between the data sources and the dashboard. Supabase as the long term store, pgvector for unstructured stuff like contracts and reviews, n8n as the orchestration layer. Six core domains every seller shares (orders, products, analytics, ads, finance, affiliates and creators) plus an optional documents layer. He closes with a dual write migration pattern so you can flip between old and new without taking the business offline.
Then Danny turns to video and voice. Remotion looks like toy town out of the box, but with the right plugins (motion blur, transitions, captions, shapes, fonts, rendering) and Claude doing the orchestration, it becomes a serious production tool that can pull in your footage, branding and design system. On the voice side, he has tested VoiceBox and F5TTS and come back to ElevenLabs Multilingual v2 at £22 a month. The missing gap everywhere is cadence. He also names the deeper bet: as the market floods with AI generated content, authentic voice becomes the differentiator that cannot be cloned.
Key Topics
Why dashboards lie when wired straight into Amazon, TikTok and Shopify
The "brain" pattern: Supabase, pgvector and n8n as a centralised data layer
The six core data domains every seller needs (plus a 7th for documents)
Dual write migration so the old system and brain run in parallel
Remotion as a code based video tool, and what it needs to stop looking toy town
The four layer creative workflow: brief, story skeleton, treatment, scene by scene
ElevenLabs vs VoiceBox vs F5TTS for voice cloning your own voice
Why cadence is the last hard problem in synthetic voice
The authenticity premium in an AI flooded market
Timestamps
[00:00] Intro and welcome back to Claude Sessions
[00:34] Shubhash kicks off: where to put the data you pulled last week
[01:04] "Your dashboard is lying to you" and the polished dashboard pitfall
[02:32] Dashboard is a window. The brain is the room behind it
[04:54] Tech stack: Supabase (Postgres), pgvector, n8n
[05:54] The six fundamental data domains
[06:26] Orders, products, analytics, ads, finance, affiliates and creators
[08:30] The optional 7th layer: unstructured documents via pgvector
[09:44] Dual write pattern for safe migration
[10:48] Three takeaways: audit, list domains, build one table at a time
[12:28] Danny on Remotion: code based video and why it is toy town out of the box
[13:51] What is missing: motion blur, transitions, captions, shapes, beat detection
[14:54] The 80+ plugin packages that turn it into a real tool
[16:56] Pulling in footage, logos, design systems and free music from Pixabay
[18:30] The 4 layer creative workflow: brief, story skeleton, treatment, scenes
[21:15] Voiceovers: ElevenLabs Pro setup and why the £22 is worth it
[22:12] VoiceBox and F5TTS field test: garbage and 5 rounds of tuning later
[23:22] Why cadence is the hardest thing for AI voice to fake
[25:42] How much reference audio you actually need (30 min min, 2 hours ideal)
[27:25] ElevenLabs UI parameters: speed, stability, similarity, exaggeration
[28:52] The authenticity premium when the market floods with AI
[30:30] Key takeaways, ElevenLabs API usage and locking in your voice once
[34:24] Aside: "insane" and "most" as the new AI tells
[36:31] SSL 2026 wrap, 18 days out, Ritu returns next week with Japan
Key Takeaways
Build a brain, not just a window. A dashboard wired straight to Amazon, TikTok or Shopify has no memory. When a feed silently fails, the dashboard happily lies. Sit a Supabase + pgvector + n8n layer in between, and your dashboard becomes a view on top of a real source of truth.
Six domains cover almost every seller. Orders, products, analytics, ads, finance, and affiliates / creators. Map every place each one currently lives, then consolidate one domain at a time. Start with one table (orders) and let Claude do the heavy lifting.
Use dual write when migrating. Write to the old store and the new brain in parallel for a week. Compare. Flip the dashboard's read side via a feature flag. If something breaks, flip back. Zero downtime, zero fear.
Remotion is a system, not a tool. Out of the box it is bare. Add the plugins (motion blur, transitions, captions, fonts, rendering), bring your own footage and design system, and let Claude orchestrate the four layer workflow: brief, story skeleton, treatment, scene by scene.
ElevenLabs Multilingual v2 still wins for voice cloning. VoiceBox and F5TTS were not close. Pay the £22, use Model 2, feed it 30 minutes minimum (2 hours ideal) of clean reference audio, and lock the setup in once.
Cadence is the last mile. AI can match tone and timbre. It still cannot match the rises, falls and micro pauses that make a sentence sound like you. Use scripts split into short paragraphs, generate three variants, and tune the language you use to talk to Claude until the cadence lands.
Authenticity becomes the moat. As written, visual and audio AI floods every channel, the brand voice that is unmistakably human becomes the differentiator. Do not give that away to save 22 dollars a month on a podcast.
Notable Quotes
"Dashboard is a window. We need a room behind the window. So the brain is going to be the room behind this window."
Shubhash Sharma
"If any of our SaaS went offline tomorrow, will our business still have its memory? The answer is no, because we haven't stored it. All we have is rented attention."
Shubhash Sharma
"When you migrate to your brain, don't rip out your old system. Use dual write. Run them in parallel for a week. If something breaks, flip it back. Zero downtime, zero fear."
Shubhash Sharma
"Remotion out of the box isn't great. It's almost like building some slides, just one step up. You have to build it as a system of what you need."
Danny McMillan
"The hardest part for AI to represent is cadence. It can get the tone of your voice. That's the easy bit. But the speed and the up and down of how you talk, that's where these models still fail."
Danny McMillan
"In our rush to use AI, you've got to remember the market floods with it. When everything sounds like AI, the only thing left is the authentic voice for your brand."
Danny McMillan
Resources Mentioned
Supabase : Postgres backend used as the long term data store for the seller "brain"
pgvector : Postgres extension for semantic search over unstructured data (contracts, reviews, supplier emails)
n8n : Orchestration layer for scheduled pulls and cron jobs with a UI
Amazon Selling Partner API (SP API) : Source for orders, inventory and finance data (covered in last episode)
Amazon Ads API : Source for ad spend, campaign and keyword data
Remotion : Code based, React powered video creation framework
ElevenLabs : Voice cloning and text to speech. Model used: Multilingual v2 (Pro plan, £22 / month)
F5 TTS : Open source text to speech model tested for voice cloning
VoiceBox by Jamie Pine : GitHub voice cloning desktop app tested by Danny
Pixabay : Free music and sound effects used inside the Remotion workflow
Loom : Source of clean voice reference audio if you record team walkthroughs
Seller Sessions Live 2026 : Conference 9 May 2026, 18 days out at recording
Hosts
Danny McMillan : Host of Seller Sessions and Claude Sessions, founder of DataBrill, building AI native tooling for Amazon sellers.
Website: https://sellersessions.com
LinkedIn: https://www.linkedin.com/in/dannymcmillan
Shubhash Sharma : Engineer building data infrastructure for Amazon and TikTok Shop sellers. Returning Claude Sessions co host.
What's Next
Next week: Ritu returns from Japan with three subjects covered in this month's rotation.
In 18 days: Seller Sessions Live 2026 in London on 9 May. Modular format, new venue confirmed.
About Seller Sessions
Seller Sessions is the leading podcast for serious Amazon sellers, hosted by Danny McMillan since 2017. Claude Sessions is the AI focused monthly strand where Danny and rotating co hosts work through the practical wins, false starts and engineering reality of building with Claude, MCPs and the wider AI stack inside real seller businesses.