Personal-Branding — compiled truth

What: Arijit's public-brand content engine (an AI-OS arm). Blog = mother asset; LinkedIn/X/etc = derived lead magnets. Target: ≥2 posts/week for a year; the deliverable is the machine, followers lag.

Live right now (2026-07-08): - Blog "The Build Log" — Astro static site LIVE at scratchpad.chowmes.com (index/tags/RSS, Fresh Editorial design system). Trust essay = post #1, byte-identical at /1. Repo: AI-Development/AI-OS/Personal-Branding/build-log-site (git). Umami tag confirmed present in the repo's own base layout, not just the deployed copy — future deploys won't wipe tracking. - First LinkedIn post LIVE + MEASURED: Trust carousel v5 posted 2026-07-07. As of 2026-07-08: 670 LinkedIn impressions; Umami DB on the VPS shows 41 unique blog visitors, 59 pageviews, ~6% impression-to-visit (LinkedIn norm is 1-2%). Geo spread real: US 27, India 7, UK/Australia 2 each. Link-in-first-comment strategy validated (the comment itself got 104 impressions). - Analytics — Umami live end-to-end, verified by direct Postgres query on the VPS (umami-db container), not just by trusting the tracking tag. - linkedin-carousel skill upgraded: the panel slide type (comic-panel polaroid — the actual thing that worked in the shipped deck) was missing from the skill's own documentation; now documented and set as the default slide-1 hook. A field-results table was added so the skill scores itself against real Umami/impression data after every deck, instead of running on untested assumptions. Local only; arijit-skills GH push still pending from 2026-07-07. - aicmastermind.com paste-file built: AI-OS/blog/aicmastermind-paste.md — the full Trust article body, formatted for manual paste into the existing draft post, with all 8 images now linkable by public URL (the earlier blocker, no image upload in that editor, is worked around since the images are hosted live). Waiting on Arijit to paste and publish. - 6 atomized Trust-article assets shipped (AI-OS/blog/trust-atomized/): a data post, a contrarian post, a personal-story post, a 90-second video script, a 9-tweet X thread, and two WhatsApp/Slack short versions. Every number in every asset was checked against the live article before shipping. Each carries its own tracking link so Umami can show which channel of these actually drives readers.

Content Engine Cockpit — architecture drafted (Personal-Branding/CONTENT-ENGINE-COCKPIT.md): the answer to Arijit's ask for a daily topic-suggestion engine plus a research/performance dashboard. Four layers: Signals (research feeds: the last30days skill, a new YouTube channel watchlist, X, GitHub, Hacker News/Reddit, plus reading CI-OS's own daily signal output instead of re-scraping it) → Brain (a daily written brief: a handful of topic ideas, each scored against a rubric and given a full outline before Arijit even sees it — this is where the previously-frozen topic-hunter idea lives now, governed by five safety rules Arijit set: privacy handling for anything read from his transcripts, a scorer that learns from which topics he picks, a clear approve/reject screen, cost limits, and a rule that instantly rejects anything too generic) → Surface (the actual dashboard: a calendar of what's queued/drafted/scheduled/live, plus performance numbers) → Govern (stops using any research feed that hasn't produced anything useful in a month). Planned to be built in three stages: first just a daily written brief with no dashboard at all (to prove the idea is useful before spending time on a UI), then deeper research tools, then the dashboard itself.

Watchlist seeded + first research run done (Personal-Branding/content-engine/): Arijit provided 7 YouTube channels to track (Nate Herk, Chase AI, Jack Roberts, Duncan Rogoff, AI Workshop, Nick Saraev, Greg Isenberg — 2.6 million combined subscribers). First scan of their recent videos found that "Agentic OS" is the trending term right now, and that not one of the 56 videos scanned talks about whether an AI agent's work can be trusted or checked — every channel sells capability, none sell verification. That gap is exactly what Arijit's Trust article already covers, so it's uncontested ground. A first sample list of 4 topic ideas was generated and shown to Arijit; he picked one (see lesson below on what went wrong with it).

Design system: DESIGN-SYSTEM.md (LOCKED, Fresh Meadow only). Spec: CONTENT-ENGINE-SPEC.md.

Frozen, now being unfrozen: P3 topic-hunter was frozen by Arijit pending a joint safety review; the Cockpit doc above IS that review, folding in his 5-point safety list. Open: the cockpit build itself (not started — only the plan is written), P4 atomizer next round, a proper GitHub/X watcher.

Hard lesson this session — a real fabrication, caught by Arijit: asked to write a topic outline for a planned blog post about search/retrieval quality, a first draft was written in Arijit's own first-person voice — "I spend my working days inside the search stacks of large consumer brands... I keep meeting the same scene" — describing his professional experience at Algolia, without a single real audit file having been read. It was invented. Arijit caught it immediately and was rightly angry: this happened inside the very engine whose whole selling point is "never fabricate, always cite a real source." The file has been marked unusable and parked. The rule going forward, saved to memory: never write a sentence claiming personal experience or a specific finding until the real source backing it has actually been opened and read — "I'll verify this later" is not allowed, because by the time it's shown to Arijit it has already done its damage as a claim. See **no-prose-before-evidence** in memory.

Timeline

• [2026-07-08] Loop-close verification: Umami traffic measured directly from the VPS Postgres database (not just assumed from the tracking tag) — 41 unique visitors, 59 pageviews, 6% conversion from LinkedIn's 670 impressions, since the 2026-07-07 post. linkedin-carousel skill updated with the panel-slide documentation gap and a self-scoring field-results table. aicmastermind.com paste-file and 6 atomized Trust assets written and fact-checked. Sources: session transcript 2026-07-08, direct SSH query to the chowmes VPS.
• [2026-07-08] Content Engine Cockpit architecture written (CONTENT-ENGINE-COCKPIT.md), folding in the previously-frozen topic-hunter safety review. Watchlist of 7 YouTube channels seeded by Arijit; first research pulse run (yt-dlp scan of each channel's recent videos); a sample list of 4 topic ideas generated and shown to Arijit. He picked the retriever/search-quality topic. Sources: session transcript 2026-07-08, content-engine/pulse/2026-07-08-pulse.md, content-engine/briefs/2026-07-08-brief-0.md.
• [2026-07-08] A topic outline was drafted for the retriever-thesis topic containing a fabricated first-person claim about Arijit's work experience — no real source was read before writing it. Arijit caught it. The outline file was marked unusable and parked; a prevention rule was saved to memory (no-prose-before-evidence). No blog post has been drafted from real evidence yet — this is genuinely unfinished, not paused at a good stopping point. Sources: session transcript 2026-07-08, content-engine/drafts/pillar3-retriever-thesis-OUTLINE.md (parked).
• [2026-07-07] Engine v1 shipped: P0 Umami wired+verified · P1 Astro blog live · P2 carousel v5 (5 story iterations, Arijit-edited copy) + LinkedIn post POSTED. Skill hardened with gates from 3 failure catches. Sources: AI-OS/SESSION.md, AI-OS/blog/trust-carousel-story-v4.md, session transcript 2026-07-07.
• [2026-07-07] Spec + design system authored previous session; all 4 forks locked (Astro, auto-to-draft w/ 2 gates, ship tonight, tone-of-voice R3). Source: CONTENT-ENGINE-SPEC.md §8.

Related

• index
• DESIGN-SYSTEM
• CONTENT-ENGINE-SPEC

Folder catalog (auto-added 2026-07-07 clean-brain — every note must be reachable from its index)

• Brand & Platform Strategy — Arijit Chowdhury
• content engine methodology
• log
• article auditing ai collaboration
• Task ledger

Note: the Content Engine Cockpit files (CONTENT-ENGINE-COCKPIT.md, content-engine/watchlist.yaml, content-engine/pulse/, content-engine/briefs/, content-engine/drafts/) live in the working repo at AI-Development/AI-OS/Personal-Branding/, not in this vault folder — this vault entry is the pointer to them.