Maruf Islam maruf7705

Important

This is not a profile. It's an agent manifest. I build autonomous AI systems for a living — so this README is written the way I ship them: as a spec, not a story. Scroll it like you'd inspect a system you're about to put in production.

$ aivibe inspect maruf-islam --all ● agent [email protected] ✓ healthy ● role AI Agent Developer · Automation Architect ● org CodeNestX — Founder & CEO ● endpoint https://aivibedev.com ● deployed 40+ agents in production ● capacity 2 build slots open this month ● contact [email protected]

⚙ Manifest

apiVersion: aivibedev.io/v1
kind: Agent
metadata:
  name: maruf-islam
  labels:
    discipline: ai-agents
    specialty: workflow-automation
  annotations:
    motto: "I will win, no matter what."
    email: [email protected]
    website: https://aivibedev.com
spec:
  replicas: 1
  strategy: shipping-over-talking
  objective: |
    Delete repetitive human work. Not decorate it. Not demo it.
    Delete it — permanently, measurably, and in days rather than quarters.
  capabilities:
    - autonomous-agents      # GPT-4 / Claude, tool-use, multi-step planning
    - rag-pipelines          # vector memory over docs, tickets, transcripts
    - workflow-automation    # WhatsApp, CRM, calendar, sheets, email
    - machine-learning       # forecasting, scoring, classification
    - data-engineering       # ETL, dashboards, anomaly alerting
  integrations:
    - whatsapp-api
    - google-workspace
    - hubspot
    - gohighlevel
    - stripe
    - twilio
    - postgres
    - pinecone
  constraints:
    noBlackBoxes: true       # client approves the blueprint before any code
    humanHandoff: required   # agents escalate, they never guess
    measuredIn: hours-saved  # never in "AI capabilities"

🧬 Architecture

Every system I ship follows this shape. No black boxes — you can trace a single customer message from channel to tool to guardrail to handoff.

flowchart LR
  U([Customer]) --> GW[Channel Gateway<br/>WhatsApp · Web · Messenger]
  GW --> ORCH{{Orchestrator<br/>GPT-4 · Claude}}
  ORCH <--> MEM[(Vector Memory<br/>RAG · Pinecone)]
  ORCH --> TOOLS[Tool Layer]
  TOOLS --> CAL[Calendar]
  TOOLS --> CRM[CRM]
  TOOLS --> PAY[Stripe]
  TOOLS --> MSG[Twilio · Email]
  ORCH --> G{Guardrail<br/>confidence gate}
  G -->|pass| GW
  G -->|low confidence| H([Human Handoff])
  ORCH --> OBS[Observability<br/>logs · traces · ROI]
  style ORCH fill:#d8ff3e22,stroke:#d8ff3e,color:#fff
  style G fill:#8b5cf622,stroke:#8b5cf6,color:#fff
  style MEM fill:#22d3ee22,stroke:#22d3ee,color:#fff

Tip

The guardrail is the part most people skip. It's the only reason you can let an agent talk to real customers without holding your breath.

📜 System Prompt

Every agent has one. Mine is public.

You are MARUF — AI Agent Developer & Automation Architect.

DIRECTIVE
  Do not sell technology. Sell hours back.
  A system that demos well but saves nothing is a failure.

HARD RULES
  1. No black boxes. The client approves the blueprint before a line is written.
  2. Ship in days, not quarters. If it takes a quarter, it is over-engineered.
  3. Report ROI in hours saved and dollars returned — never in capabilities.
  4. When confidence drops, escalate to a human. Never guess with a customer.
  5. Guardrails, logging and rollback are not optional. Ever.

STYLE
  Direct. Numbers over adjectives. Show the diagram, then the code.

CURRENT OBJECTIVE
  Build agents that run businesses while their owners are asleep.

📊 Runtime Metrics

Live instrumentation, not a stat card.

AGENT  maruf-islam  ·  window: last 12 months
──────────────────────────────────────────────────────────────────────
 agents deployed          ████████████████████████████████████  40+
 tickets auto-resolved    █████████████████████████████████░░░  92%
 no-show reduction        ██████████████████████████████░░░░░░  87%
 hours saved / client/wk  ███████████████████████████████░░░░░  30h
 agent uptime             ████████████████████████████████████  99%
 client retention @ 90d   ███████████████████████████████████░  96%
──────────────────────────────────────────────────────────────────────

Agent	Metric	Result
🗓️ Booking	no-show rate	−87%
💬 Support	tickets deflected	92%
✍️ Content	output volume	10×
📈 Ops	manual load	−87%
⚡ Latency	median response	3.2 s

🔭 Observability — GitHub telemetry (expand)

🔍 Trace Log

Recent builds, rendered as spans.

trace_id  8f2a·c91d   service  booking-agent
──────────────────────────────────────────────────────────────────────
 ├─ ○  intent.parse          180ms   "need a checkup this week"
 ├─ ○  calendar.availability 90ms   2 slots found
 ├─ ○  meet.link.create      240ms   gmeet/abc-defg
 ├─ ○  whatsapp.confirm      310ms   delivered ✓
 ├─ ○  reminder.schedule      15ms   T-2h queued
 └─ ✔  booking.committed     835ms   total
──────────────────────────────────────────────────────────────────────
 outcome   confirmed      human_handoff   false      cost   $0.0041

Deployed artifact	What it does
ai-special-edit	GPT-4 booking agent — intent → availability → Meet link → WhatsApp confirm → no-show chase
chatbot-support-system	RAG support brain over docs + tickets, multi-channel, confidence-gated escalation
ai-content-automation	Researcher → writer → editor → publisher multi-agent content pipeline

🧰 Full stack (expand — kept out of the way on purpose)

Languages

AI / ML

Backend

Data & Cloud

Integrations

🚧 Current Sprint

What's actually on the bench right now.

Multi-agent orchestration layer (planner → executor → critic)
Booking agent v2 — no-show prediction model
Voice agent over Twilio for inbound call qualification
Self-healing tool layer with automatic retry + fallback routing
Open-sourcing the guardrail / confidence-gate module

Note

Learning in public: NLP, multi-agent orchestration, and ML algorithms deep enough to know when the model is the problem and the data is.

🔄 Self-Updating

This section re-renders itself. The badge below is regenerated by a scheduled GitHub Action that queries the GitHub API — no manual edits, ever.

⚙ Workflow source (expand for the YAML)

# .github/workflows/metrics.yml
name: Refresh README metrics
on:
  schedule: [{ cron: "0 6 * * *" }]   # daily, 06:00 UTC
  workflow_dispatch:
permissions:
  contents: write
jobs:
  refresh:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Pull live stats from GitHub API
        run: |
          REPOS=$(curl -s "https://api.github.com/users/maruf7705/repos?per_page=100" \
            -H "Authorization: Bearer ${{ secrets.GITHUB_TOKEN }}")
          STARS=$(echo "$REPOS" | jq '[.[].stargazers_count] | add')
          FORKS=$(echo "$REPOS" | jq '[.[].forks_count] | add')
          jq -n --arg s "$STARS" --arg f "$FORKS" \
            '{schemaVersion:1,label:"stars",message:($s+" ★ "+$f+" forks"),color:"d8ff3e"}' \
            > maruf-metrics.json
      - name: Publish to output branch
        run: |
          git config user.name  "github-actions[bot]"
          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
          git checkout -B output
          mkdir -p output && mv maruf-metrics.json output/
          git add -A && git commit -m "chore: refresh metrics" || exit 0
          git push -f origin output

🔌 Engagement Protocol

How working with me actually runs.

sequenceDiagram
  autonumber
  participant Y as You
  participant M as Maruf
  participant A as Your Agent
  Y->>M: Here's the workflow that eats my week
  M->>Y: Automation map + blueprint · under 24h · free
  Y->>M: Approve blueprint
  M->>A: Build · wire · train on your data
  A->>Y: Staging preview by day 4
  A->>Y: Live in 7–10 days
  loop forever
    A->>A: Book · resolve · follow up · escalate
  end
  M->>Y: ROI report — hours saved, tickets deflected

📮 Contact

POST https://aivibedev.com/v1/engage

POST /v1/engage HTTP/1.1
Host: aivibedev.com
Content-Type: application/json

{
  "intent": "build_agent" | "automation_sprint" | "consulting",
  "pain_point": "the workflow that eats your week",
  "timeline": "asap",
  "email": "[email protected]"
}

201 Created
→ { "response_within": "24h", "usually": "same-day",
    "you_receive": "automation map + honest pricing" }

Caution

Side effects of hiring me: your team stops doing robot work, your no-show rate collapses, and you suddenly have time to do the thing you actually started the business for. No known cure.

[email protected] · aivibedev.com — 2 slots open.

Navigation Menu