LinkedIn Assistant

An AI-powered LinkedIn interaction bot that automatically discovers relevant posts, evaluates them against your professional profile, and delivers actionable engagement suggestions — complete with ready-to-paste comment drafts — directly to your Telegram chat.

It also serves as an SSI (Social Selling Index) optimizer: every suggestion is framed as a strategic engagement opportunity aligned to your personal brand.

---

Features

• Automated post discovery — searches LinkedIn (authenticated scrape or Tavily fallback) across configurable topic queries.
• Two-stage AI filtering — a fast Groq triage model eliminates irrelevant posts cheaply; only approved posts reach the deeper DeepSeek analysis.
• SSI strategy reports — each approved post produces a relevance score, a strategic engagement angle, and a polished comment draft.
• Seen-URL deduplication — posts already processed in previous runs are skipped, so each digest contains only new content.
• Telegram digest delivery — results are batched into Telegram messages that respect the 4096-character limit.
• Interactive bot — a long-polling Telegram bot lets you trigger runs and check status on demand.

---

Architecture Overview

index.js  (pipeline entry point)
│
├── linkedin_scraper.js   → Playwright scrape of your own profile + activity
│                           (falls back to data/cv.md when no cookie is set)
│
├── search_engine.js      → Playwright authenticated search OR Tavily API fallback
│                           Returns: [{ title, url, content }]
│
├── triage_filter.js      → Groq LLM (fast, cheap): boolean SI/NO relevance gate
│
├── ssi_analyzer.js       → DeepSeek LLM (deep): SSI report + comment draft
│
├── seen_store.js         → File-backed Set of processed URLs (data/seen_urls.json)
│
└── telegram_sender.js    → Telegram Bot API wrapper (Markdown → HTML conversion)

bot.js  (interactive Telegram bot, long-polling)
└── Dispatches /suggest → spawns index.js as a child process

Data flow

Profile scrape
      │
      ▼
Topic search (3 queries × 5 posts)
      │
      ▼
Deduplicate by URL → filter seen URLs
      │
      ▼
For each new post:
  Triage (Groq) ──REJECTED──► skip, mark seen
      │
   APPROVED
      │
      ▼
  SSI Analysis (DeepSeek)
      │
  Score ≥ MIN_MATCH_SCORE?
      │
     YES
      │
      ▼
  Append to Telegram digest

Send batched digest → Telegram

---

Setup

Prerequisites

• Node.js 20 or later
• A Chromium-compatible browser accessible to Playwright (npx playwright install chrome)
• API keys for: Groq, DeepSeek, Telegram Bot, and optionally Tavily

1. Install dependencies

npm install
npx playwright install chrome

2. Configure environment variables

Copy the example file and fill in your credentials:

cp .env.example .env

Open .env and set the following variables:

Variable	Required	Description
TELEGRAM_BOT_TOKEN	Yes	Token from @BotFather
TELEGRAM_CHAT_ID	Yes	Your personal Telegram chat ID (use @userinfobot)
GROQ_API_KEY	Yes	API key from console.groq.com
DEEPSEEK_API_KEY	Yes	API key from platform.deepseek.com
TAVILY_API_KEY	No	Required only when LINKEDIN_LI_AT is not set
LINKEDIN_LI_AT	No	LinkedIn session cookie value (enables authenticated scraping)
LINKEDIN_PROFILE_URL	No	Your LinkedIn profile URL (required if LINKEDIN_LI_AT is set)
TRIAGE_MODEL	Yes	Groq model ID, e.g. llama3-8b-8192
ANALYSIS_MODEL	Yes	DeepSeek model ID, e.g. deepseek-chat
MIN_MATCH_SCORE	No	Minimum relevance % to include a post (default: 70)

Obtaining LINKEDIN_LI_AT

1. Log in to LinkedIn in your browser.
2. Open DevTools → Application → Cookies → www.linkedin.com.
3. Copy the value of the li_at cookie.

Warning: Do not share your li_at cookie. It grants full access to your LinkedIn session.

3. (Optional) Add a profile fallback

If you do not set LINKEDIN_LI_AT, place your professional profile as plain text or Markdown in data/cv.md. The pipeline will use this file instead of scraping LinkedIn.

---

CLI Commands

Command	Description
npm start	Run the full pipeline once and send the digest to Telegram.
npm run bot	Start the interactive Telegram bot (long-polling, runs until stopped).
npm test	Run the unit test suite with Vitest.

---

Telegram Bot Commands

Once the bot is running (npm run bot), send these commands from your Telegram account:

Command	Description
/suggest	Trigger a full pipeline run. Results are sent to the same chat when complete.
/profile	Display the profile / CV currently loaded by the bot (truncated to 3000 chars).
/status	Report whether a pipeline run is currently in progress.

Only messages from the configured TELEGRAM_CHAT_ID are accepted; all other senders are ignored.

---

Project Structure

.
├── index.js              # Pipeline entry point
├── bot.js                # Interactive Telegram bot
├── src/
│   ├── config.js         # Environment variable loading
│   ├── linkedin_scraper.js  # Playwright profile & activity scrapers
│   ├── search_engine.js  # LinkedIn post search (Playwright + Tavily fallback)
│   ├── triage_filter.js  # Groq-powered relevance gate
│   ├── ssi_analyzer.js   # DeepSeek SSI strategy analyser
│   ├── seen_store.js     # File-backed deduplication store
│   └── telegram_sender.js  # Telegram Bot API wrapper
├── data/
│   ├── cv.md             # (optional) Local profile fallback
│   └── seen_urls.json    # Auto-generated; tracks processed posts
├── tests/                # Vitest unit tests
├── .env                  # Local credentials (never commit this)
├── .env.example          # Template for required environment variables
└── package.json

---

Running Tests

npm test

Tests are written with Vitest and cover the triage filter, SSI analyser, seen store, and Telegram sender modules.

---

License

MIT