LinkedIn Content Agent: AI-Powered Personal Branding

Fully automated pipeline that generates Claude AI-written LinkedIn drafts from a Google Sheet, sends them via Telegram for approval, then auto-posts with AI-generated images to LinkedIn — on a daily cron schedule.

Product Manager & Engineer — System Design, Prompt Engineering, API Integration, Automation

Next.js

Claude AI (Anthropic)

Google Sheets API

Telegram Bot API

LinkedIn API

Vercel Cron

30+ min/day

Manual time saved

Draft options/day

Integrations

1-tap approve

Human touchpoints

The Problem

Maintaining a consistent LinkedIn presence as a PM requires daily content creation — but writing, reviewing, and posting quality content manually takes 30+ minutes a day. Generic AI tools produce bland, corporate-sounding posts that don't reflect a real voice or perspective.

The Solution

Built a fully automated Next.js + Claude AI pipeline: reads topic notes from Google Sheets, generates 3 tailored post drafts (story, insight, conversation-starter) using a custom PM voice prompt, delivers them to Telegram for one-tap approval, generates a matching image via AI, and posts directly to LinkedIn — all on a daily cron schedule.

Product Artifacts

GitHub

Source code

View

Case Study

Context

Consistent LinkedIn posting is a known PM career accelerant — but the activation energy for daily content is high. Most people start, then stop within 2 weeks.
Generic AI tools (ChatGPT, Jasper) produce content that sounds like everyone else: buzzword-heavy, vague, and clearly AI-written. They don't have context about your voice, expertise, or target audience.
The real friction isn't ideation — it's the gap between having a rough thought and turning it into something worth posting. That's where most people drop off.
Target: a system that does the heavy lifting daily, surfaces polished options, and gets out of the way — requiring only one human decision: which post to use.

Decision

Used Google Sheets as the content queue — low friction for adding raw notes, no custom UI to build, easy to batch ideas in advance.
Designed a custom Claude system prompt to enforce a specific voice: PM with ML background, conversational, no em-dashes or buzzwords, under 180 words, ends with a question.
Generated 3 draft types per run (story, insight, conversation-starter) so there's always a best fit for the day's mood and context.
Telegram as the approval interface — one-tap selection from phone, no login required, fits into a morning routine in under 30 seconds.
Vercel Cron for scheduling — zero infrastructure overhead, runs daily at a fixed time, logs visible in dashboard.

Execution

Google Sheets integration reads the first 'unused' row by status column, pulls topic, raw notes, target audience, angle, and CTA fields.
Claude API (claude-sonnet-4-6) generates 3 drafts using a detailed system prompt that encodes Mohak's voice, rules, and post structure.
Telegram bot delivers formatted drafts with recommended flag highlighted — user replies with post number to select.
On selection, generates a matching AI image via image API, registers and uploads it to LinkedIn's media asset endpoint, then publishes the post.
Google Sheet row is updated to 'used' with timestamp and published text to prevent re-use and track history.

Outcome

Reduces daily LinkedIn content effort from 30+ minutes to a single Telegram tap.
Consistent voice and quality across posts — system prompt encodes years of PM perspective, not generic AI tone.
Full audit trail in Google Sheets: every post drafted, selected, and published is logged with date and text.
End-to-end automation from idea to published post with exactly one human decision point.

Learnings

The prompt is the product. The entire quality difference between bland AI posts and genuine-sounding ones lives in the system prompt — getting it right took more iteration than the code.
Approval friction matters. Telegram was the right choice over email or a web UI because it's already open in the morning. Reducing steps from 3 to 1 dramatically increases follow-through.
Google Sheets as a database is underrated for personal automation — it's the fastest possible content queue with zero schema setup and full visibility.
Designing for yourself is a great forcing function: you immediately feel every rough edge in the UX because you're the daily user.

View All Projects