Case Study

LinkedIn on Autopilot — Zero Daily Involvement

Rating: 5
Author: Lets Viz Team

A full-stack Windows desktop app that plans, writes, publishes, and engages on LinkedIn autonomously — so the founder never has to.

For a consulting firm selling expertise, LinkedIn is the primary channel for authority-building and inbound leads. But managing it well — daily posts, thoughtful comments on target accounts, performance tracking — is a part-time job in itself. Lets Viz built LV LinkedIn OS: a production Windows desktop app (Electron + React + FastAPI) that automates the complete LinkedIn lifecycle. Playwright connects to the founder's real browser session via CDP — human-like automation with no bot signals. Four autonomous loops. Zero daily involvement required.

B2B SaaS / Data & Analytics · Internal — Lets Viz Technologies·ElectronReactFastAPIPythonSQLitePlaywrightClaude CLIOpenAI

LV LinkedIn

Lets Viz Technologies

System Online

▪Dashboard

inLinkedIn

✦Content Studio

◈Intelligence

≡Content Pipeline

◉Auto-Engage

✧Comments

⚙Settings

LV LinkedIn OS

FileEditViewHelp

Engagement Activity

Comments Made

▼ 45% WoW

This Week

228

Total Comments

Profiles Engaged

▼ 43.2% WoW

Visited Today

Content Pipeline

78 total posts

CalendarKanbanListEvergreen

Ideas0

Empty

Drafts45

** Clinical labs run 500+ tests daily — the case for real-time LIS-to-Power BI pipelines

Image

** AI automation is processing lab results faster — reporting layer is a 24-hour lag

Image

Scheduled5

Who's accountable when your automated financial report is wrong — the AI or the analyst?

Image

From experimenting with AI to measurable ROI in BI

Image

Results That Speak for Themselves

Autonomous workflows: plan, publish, engage, leads

Daily involvement required from the founder

App pages — dashboard to auto-engage to lead pipeline

100%

Human-like browser automation — real session, no bot detection

From challenge to outcome

The Problem & Our Approach

A Founder Cannot Also Be a Full-Time LinkedIn Manager

For a consulting firm selling expertise, LinkedIn is the primary channel for authority-building and inbound leads. But managing it well — daily posts, thoughtful comments on target accounts, connection outreach, performance tracking — is a part-time job in itself. LinkedIn cannot be a side task for a founder who is also the delivery lead. It needs its own system — or it does not get done.

No consistent publishing cadence — posts written when there was bandwidth. Weeks passed without content. No idea pipeline, no scheduling system, no way to write at scale without consuming billable hours.
Engagement was manual or absent — building authority requires commenting on the right posts consistently. Doing this manually across dozens of target profiles every week was not sustainable.
Lead generation had no system — potential clients existed in LinkedIn search results but there was no pipeline for finding, qualifying, and nurturing them through to a conversation.
No feedback loop — no visibility into which posts drove profile views, which engagement targets converted, or what content resonated with the BI and analytics audience.

A Full-Stack Desktop App That Runs LinkedIn for You

Lets Viz designed and built LV LinkedIn OS — a production Windows desktop application (Electron + React + FastAPI) that automates the complete LinkedIn lifecycle. A Playwright-powered browser connects to the founder's real LinkedIn session via CDP, enabling human-like automation that operates within normal session behaviour. Every long-running operation runs as a background task with a live polling UI — nothing blocks, nothing times out on navigation.

Idea Engine — generates post ideas tailored to the BI/analytics niche. Per-post industry tagging (general, healthcare BI) with matching system prompts and hashtag pools. Feeds the Content Pipeline queue automatically.
Content Studio — AI-assisted drafting in the founder's voice via Claude CLI. Produces an image brief; user triggers image generation (OpenAI GPT-Image-2 primary / Gemini fallback) on demand after reviewing text.
Publisher — scheduled auto-posting at optimal times. Posts queue with status tracking (ready → publishing → published / failed). Background task with live status poll.
Auto-Engage — browses target profiles, scrapes recent posts from /recent-activity/all/, posts AI-generated comments specific to the post content. Human-like timing: log-normal delays, mouse jitter, periodic detours. Auto-rejects inactive profiles (>60 days); skips profiles with no recent posts (>7 days).
Performance Scraper — visits each post URL individually to collect impressions (owner-only stat). Drives the Karpathy scoring loop: high-performing post formats get higher queue weight.
Lead Pipeline — scrapes LinkedIn search results for potential leads, manages them through a Kanban board (New → Contacted → Replied → Meeting → Converted), feeds qualified leads into connection campaigns with AI-personalized outreach notes.
Google Sheets Sync — incremental bidirectional sync. SQLite trigger marks records dirty on change; sync only pushes dirty rows — never full-table overwrites.

What We Achieved

4 autonomous loops in production: content pipeline, scheduled publishing, overnight auto-engagement, lead pipeline

0 daily involvement required — a week's content reviewed and queued in one sitting

Engagement happens overnight — founder wakes to a comment history, not a to-do list

Karpathy scoring loop closes the feedback cycle: scrape impressions → score formats → weight queue toward what works

30+ versioned SQLite migrations — schema evolves without data loss

Platform change caught automatically — LinkedIn's May 2026 timestamp removal detected and handled with no manual intervention

8 app pages covering every LinkedIn workflow: Dashboard, LinkedIn session, Content Studio, Intelligence, Content Pipeline, Auto-Engage, Comments, Settings

Eight Modules — Complete LinkedIn Lifecycle

Module	What It Does
Idea Engine	Niche-matched post ideas, per-industry tagging, hashtag pools. Auto-feeds Content Pipeline.
Content Studio	Claude CLI drafting in founder voice. Image brief + on-demand GPT-Image-2 generation.
Publisher	Scheduled auto-posting. Status tracking: ready → publishing → published / failed.
Auto-Engage	Browses targets, scrapes recent posts, posts specific AI comments. Log-normal timing, mouse jitter. Skips stale profiles.
Performance Scraper	Visits each post URL for impressions. Feeds Karpathy scoring loop — better formats get higher queue weight.
Feed Intel	Scrolls feed, ranks top 30 posts by reaction+comment signal. Surfaces in Feed Intel tab.
Lead Pipeline	LinkedIn search scraper. Kanban board (New → Contacted → Replied → Meeting → Converted). AI outreach notes.
Google Sheets Sync	Incremental bidirectional sync via SQLite dirty-flag pattern. No full-table overwrites.

“

The app behaves like a dedicated LinkedIn manager who starts work at 2 AM, never misses a schedule, and leaves comments that read like a human wrote them.

Lets Viz Team·Internal — LV LinkedIn OS

Human-Like Automation That Does Not Get You Banned

LinkedIn's anti-automation detection is behavioural — it looks for inhuman timing, identical mouse paths, and activity patterns that no real user would produce. LV LinkedIn OS was designed from the ground up to look and behave like a person.

Log-normal timing between actions — natural variance, not fixed intervals.
Mouse jitter and variable scroll depth on every page navigation.
Periodic detours to the home feed and network page every 5–7 profiles.
Accepts 1–2 connection requests during detour pauses.
CDP connects to the user's live Electron BrowserView — same cookies, same session, same IP as a normal browser.
All comments are generated per-post via Claude CLI — specific to the content, not templated.
The session looks human because it is human — same browser, same account, same session.

The Feedback Loop That Makes the Content Smarter Over Time

Most LinkedIn automation tools post and forget. LV LinkedIn OS closes the loop: every published post's impressions, reactions, and comments are scraped back into the app and fed into a scoring model that adjusts which content formats get prioritised in the queue. The more the system runs, the better it gets at predicting what performs for this specific audience.

Performance scraper visits each post URL individually to collect impressions — an owner-only stat not visible in feed listings.
Feed signal score uses a 48-hour recency window and degrades gracefully when the feed_observations table is empty.
Repost detection uses /in/ URL slug comparison — first-name matching was too fragile and was replaced.
Post age parser checks 'mo' before 'm' in regex to prevent '5 months' matching as '5 minutes' — a subtle but real parsing trap.
LinkedIn removed <time datetime> tags from profile Activity previews in May 2026. The app detects this automatically and falls back to full post scrape — logged to ~/.lv-linkedin/backend.log for future DOM-change diagnosis.

The Pattern — Autonomous Presence Management

LV LinkedIn OS is one instance of a broader capability: taking a high-frequency, expertise-dependent workflow and replacing it with an instrumented, agent-driven system that a small team can monitor from a single interface. The same architecture — background tasks, scoring loops, human approval gates, and adaptive feedback — applies wherever expert throughput is the bottleneck.

Built for real production use: signed NSIS installer, persistent local SQLite DB at ~/.lv-linkedin/, live session automation, long-running async ops.
Adaptive by design: scoring loops and per-post industry tagging mean the system gets better the longer it runs.
Platform-resilient: DOM change detection and fallback paths mean a LinkedIn UI update does not silently break the automation.
Human-in-the-loop where it matters: every draft is reviewable before publish; every engagement run is logged with full audit trail.
If your founder is also your marketing department, something has to change. Lets Viz builds the system that runs it for you.

Discuss a similar B2B SaaS / Data & Analytics project

Lets Viz runs a paid discovery audit ($500–1,000, credited toward the project) to scope your requirements, data model, and architecture before writing a line of code.

From Lets Viz

Helping companies make data-driven decisions to achieve their Goals

Experience superior data analysis with our custom dashboards. Specialising in all tools and technologies, we turn data into visually compelling, actionable insights.

NDA-safe · No obligation · Clear next steps

Get in Touch Live Dashboard Experience