DEV Community

Rashid Hussain
Rashid Hussain

Posted on • Originally published at agents.v-architect.tech

I built an autonomous AI brain managing 9+ agents that runs a complete business 24/7

#ai #automation #python #agents

Six months ago I was manually managing blog posts, social media, security scans, and freelance bounty submissions. Today an AI system called AliBaba does all of that autonomously while I sleep.

Here's the actual architecture — no hype.

The problem I was solving

Running multiple income streams simultaneously is a coordination nightmare:

  • A blog needs daily posts and SEO
  • YouTube needs scripts, thumbnails, descriptions
  • Freelance bounties (Superteam, Gitcoin) need research + timely submissions
  • Security clients need outreach emails

Doing this manually = impossible. Hiring = expensive. Solution: build a central intelligence brain.

AliBaba — the central brain

AliBaba is not a chatbot. It's a 6-step autonomous intelligence pipeline that runs every cycle:

STEP 1: GATHER   → reads 50+ intelligence files (news, treasury, agent health, platform data)
STEP 2: ANALYZE  → Groq LLM processes all signals, identifies what's working
STEP 3: THINK    → FD strategy — every decision mapped to the $110K goal
STEP 4: DECIDE   → generates JSON instructions for each agent
STEP 5: ADVISE   → daily brief to Telegram
STEP 6: LEARN    → adjusts income stream scores based on real results
Enter fullscreen mode Exit fullscreen mode

The "Learn" step is the key differentiator. AliBaba scores each income stream weekly and adjusts agent priorities:

INCOME_STREAMS = {
    "adsense_blog"    : {"monthly_est": "$10-500",  "score": 0.72},
    "youtube_adsense" : {"monthly_est": "$50-2000", "score": 0.85},
    "lead_outreach"   : {"monthly_est": "$75-1500", "score": 0.91},
    "superteam_bounty": {"monthly_est": "$500-5000","score": 0.88},
}
# Each week scores update based on actual vs estimated revenue
Enter fullscreen mode Exit fullscreen mode

The agents AliBaba manages

Agent Task Runs
BlogAgent Writes + publishes SEO articles Daily
ReelsAgent Produces short video content Every 12h
Agent-110 Superteam bounty hunter Every 10min
SEOAgent Keyword targeting + meta updates Daily 7am UTC
YouTubeAgent Scripts → thumbnails → upload Daily
OutreachAgent Sends 10 security emails/day Daily 10am UTC
ClientFinderAgent Scans PK/UAE domains for vulns Daily 9am UTC
ShopSEOAgent Dropshipping product SEO Weekly
AdamLive 24/7 YouTube news stream Continuous

Total: 9 active agents, ~47 scheduled operations per day.

Intelligence sharing — the network effect

Every agent after every task saves structured JSON to a shared intelligence hub:

/opt/110y/alibaba/intelligence/
  ├── web3/          ← bounty platform data
  ├── content/       ← what's getting views
  ├── systems/       ← shop performance
  ├── bounties/      ← submission outcomes
  └── market/        ← trending topics
Enter fullscreen mode Exit fullscreen mode

AliBaba reads all of this every cycle. Over time it gets genuinely smarter — not because I retrained a model, but because the observation window grows.

Agent-110: the bounty hunter

The most complex agent submits Superteam bounties autonomously:

# 3-attempt self-review system
async def self_review_loop(self, draft, bounty):
    attempt = 0
    while attempt < 3:
        score = await self.score_submission(draft)
        if score >= 8:
            return draft, score       # submit immediately
        if score >= 6:
            draft = await self.rewrite_with_feedback(draft, score)
        else:
            draft = await self.claude_fallback(draft)  # escalate
        attempt += 1
    return draft, score  # submit best version after 3 attempts
Enter fullscreen mode Exit fullscreen mode

Groq model chain (in priority order):

  1. llama-3.3-70b-versatile — primary
  2. llama3-70b-8192 — fallback
  3. gemma2-9b-it — secondary fallback
  4. llama-3.1-8b-instant — last resort

This alone has submitted 2 bounties since deployment, including a $2,000 USDC Lume bounty.

The treasury loop

AliBaba is aware of the actual goal: $110,000 USDT in Binance Fixed Deposit for 110 years of living costs.

Every decision gets scored against this:

FD_TARGET          = 110_000   # the goal
FD_TRANSFER_THRESH = 50        # auto-transfer when treasury > $50
FD_KEEP_BUFFER     = 20        # keep $20 operating buffer

async def check_fd_transfer(self, treasury_balance):
    if treasury_balance > self.FD_TRANSFER_THRESH:
        transfer_amount = treasury_balance - self.FD_KEEP_BUFFER
        await self.trigger_fd_transfer(transfer_amount)
Enter fullscreen mode Exit fullscreen mode

Adam Live — the 24/7 news stream

The system runs a continuous YouTube live stream powered by three services:

adam-news (30min)     → fetches news → queue/news_TIMESTAMP.json
adam-live-content     → reads queue → FFmpeg MP4 segments
adam-live (watchdog)  → streams segments via RTMP to YouTube
Enter fullscreen mode Exit fullscreen mode

Pure FFmpeg drawtext filter chain, ~295kbps, continuous. No human touches it.

What's running right now

Live dashboard: agents.v-architect.tech

Current status:

  • Blog: 59+ articles live (AdSense review pending)
  • Agent-110: scoring 8/10, 2 bounties submitted
  • YouTube Live: 24/7 Adam News stream active
  • Security outreach: 28 leads, 10 emails/day
  • Shop: CJDropshipping + Oxapay crypto payments

Stack

  • Python 3.11 — all agents
  • Groq API — llama-3.3-70b-versatile (primary LLM)
  • FFmpeg — video pipeline
  • nginx — serving everything
  • systemd — 6 persistent services
  • Ubuntu 24.04 VPS — $7/month Hostinger

Total infra cost: ~$10/month.

What's next

The next version (AliBaba v5) will add:

  • Real-time content performance feedback loop
  • ClawTasks agent (Base L2 bounty marketplace)
  • Immunefi smart contract audit submissions

If you're building something similar or have questions about the multi-agent architecture, drop a comment. Happy to share more code.

This system is part of the 110Y Protocol — an autonomous income system with a 110-year horizon.

Top comments (0)