Unified omnichannel (voice + message) AI customer support.
MCP-native, speech-to-speech support product with BYOK across any LLM, deployable on-prem for data residency and regulated, high-scale teams. Contributed backend/automation and frontend.
Lets any LLM client operate a self-hosted WordPress site.
Remote MCP tool server exposing draft / publish / update / search actions so any MCP client (including Claude) can perform multi-step tasks on a WordPress site — no plugins, authenticated via Application Passwords, deployed on Prefect Horizon.
Gives LLM agents live GA4 + Ahrefs data as callable tools.
MCP tool server that collects GA4 analytics and Ahrefs SEO data via API and exposes it as callable tools — giving LLM agents on-demand access to live traffic and SEO metrics for retrieval and analysis.
AI agent that audits 52 SEO checks and fixes WordPress and Next.js sites on its own.
Runs a 52-check audit based on the HOTH SEO report and fixes meta titles and descriptions, headings, body copy, thin content, image alt text, WebPage schema, and internal links. Ships as the iBOS SEO Agent WordPress plugin (v1.8.0) and the @ibos/seo-agent package for Next.js. Skips page-builder pages for content, heading, and link changes, and can send every fix as a human-approved proposal. Optional GEO fixes add llms.txt, Organization schema, and answer and FAQ blocks. An off-page agent handles backlink outreach and tracks backlinks through Ahrefs and Google Search Console.
Desktop control tower that turns Claude Code, Codex, and OpenCode into scheduled, self-governing content agents.
Electron desktop platform orchestrating multiple agent CLIs via real PTY sessions and headless autopilot runs. Custom MCP server (raw JSON-RPC, zero dependencies) streams permission requests to the UI with a 10-minute deny-by-default timeout. Governed pipeline: research → create → deduplicate → QC → gated publish, enforced by a single non-bypassable chokepoint. Multi-layer duplicate detection (SHA-256, pHash, source windows, optional semantic embeddings), RAG-style retrieval of past high-performing posts, and scheduled Facebook Graph API publishing of static and OpenMontage-generated video with 6-hour performance polling stored as time-series. ~11K lines across Electron/Node.js, Python, and PostgreSQL.
Turns a written script into a finished vertical video from a single Claude Code prompt.
Claude Code workflow that takes a script to a finished 25–90 second reel, with review points at the voice, look, and final cut. Clones a Bangla narrator voice with the Gemini API and reuses it across videos; every line is checked with faster-whisper speech-to-text and re-voiced when words are missing or wrong. Real product screens are captured with Playwright (personal data blurred) so each claim is shown on a live screen, then combined with Google Flow clips, brand logos, Bangla captions, and music. Used for Managerium, PeopleDesk, TrackForce, and Travilo product reels and the Life Lesson BD series.
Claude Code
Gemini API
Google Flow
Playwright
ffmpeg
faster-whisper
Python
Node.js
Internal — details on request
Automation · n8n
Meta Ads Lead-Capture & CRM Automation
Webhook-driven lead management across all company Meta pages.
Verifies Meta's handshake, ACKs instantly, fetches lead data from the Graph API, normalizes to Dhaka time, dedups, and pushes each lead into Google Sheets and the internal CRM — with retry/fallback on every external call.
n8n
Graph API
Google Sheets
CRM
Internal — details on request
Automation · n8n
AI-Powered Daily Ads Reporting Digest
Multi-source ads digest with a hallucination guard.
Scheduled Meta + Google Ads digest that fetches live spend/CPL/CPA, runs DeepSeek analysis with strict anomaly thresholds, and auto-emails an executive summary — guarded by a verification check that flags any run not backed by real tool data.
n8n
Meta Ads API
Google Ads API
DeepSeek
Gmail API
Internal — details on request
Data · Retrieval
Leads Scraper
Sheet-triggered Google Maps lead engine, crash-safe.
FastAPI service that polls a Google Sheet, scrapes Google Maps, and enriches emails/socials from business sites — persisting each lead on finalize (crash-safe) with DB-level dedup via a UNIQUE key (domain → phone → name+address).
Every lead submitted through a Meta lead-gen form is verified, deduplicated, and pushed into the company CRM automatically. Before this pipeline, someone had to check each lead, find it, and upload it into the CRM one by one.
n8n
Meta Graph API
Google Sheets
Managerium CRM
Webhooks
Stack
What I build with.
AI / LLM
Anthropic Claude
Claude Code
OpenAI Codex
OpenCode
LLM orchestration
Autonomous agents
Agentic systems
Tool use
Function calling
RAG
Embeddings
Vector similarity
Cosine similarity
Semantic search
Prompt engineering
Context injection
Hallucination reduction
AI Agents & MCP
Model Context Protocol (MCP)
JSON-RPC
Custom MCP servers
Multi-step agents
Tool-calling
Supervised autonomy
Human-in-the-loop
Scheduling/cron
Time-series analytics
Perceptual hashing (pHash)
SHA-256
Backend & APIs
Python
FastAPI
SQLAlchemy (async)
REST APIs
Graph / Ads APIs
Webhooks
JWT auth
WebSockets
Automation
n8n
Webhook pipelines
Retry / fallback
Error handling
CRM integration
Audit logging
Data & DevOps
PostgreSQL + pgvector
Playwright
Pandas
Docker
docker-compose
Git / GitHub
Frontend
Next.js (App Router)
React
TypeScript
Tailwind CSS
About
Who's behind the work.
I'm an AI Automation Engineer who ships production systems, not just drag-and-drop flows. I build full-stack AI platforms on FastAPI and Next.js, custom MCP servers that expose real tools to LLM clients, RAG pipelines over pgvector, and end-to-end automation on n8n.
My focus is reliability: dedup and idempotency, secure credential handling, retry/fallback on every external call, and verification layers that catch when an AI answer isn't grounded in real data. I care about automations that measurably remove manual work from business teams.
Currently building AI products at AKIJ iBOS — including ChannelSync, an MCP-native omnichannel support platform — while shipping my own tools: multi-tenant RAG apps, MCP tool servers, and lead-management pipelines.
AKIJ iBOS
Dhaka, Bangladesh
Associate Developer
Aug 2026 — Present
▹Contributed backend/automation and frontend to ChannelSync — a unified omnichannel (voice + message) customer-support product: speech-to-speech, MCP-native, BYOK across OpenAI, Claude, and other LLMs.
▹Built AI automation on n8n for real business processes across marketing and other teams — automating lead handling, reporting, and publishing end-to-end, replacing hours of manual work with self-running pipelines.
▹Integrated LLM outputs with live business tools (CRM, Google Sheets, Gmail) via REST/Graph APIs, with error logging, output validation, retry/fallback, and reusable integration components.
AI & Automation Engineer (Intern)
May 2026 — Jul 2026
▹Joined as an intern, then promoted to Associate Developer on 1 August 2026 after working on ChannelSync and n8n automation projects across marketing and other teams.
ZNRF University of Management Sciences, Dhaka
B.Eng. in Computer Science & Engineering
Jul 2022 — Jul 2026
Top 5% of batch
ZNRF Trust Scholarship recipient
Activities
Leadership & Clubs
▹Ex-General Secretary — ZUMS Student Forum (2025)
▹Ex-Office Secretary — ZUMS Biz-Tech Club
▹Ex-Secretary — ZUMS English Club
▹Executive Member — BASIS Student Forum ZUMS Chapter
▹Participant — National Conference on Intellectual Property & Commercialization (ICIPC) 2024
Contact
Let's build something reliable.
Open to remote AI and automation roles, and to freelance builds — RAG systems, AI agents, MCP servers, and workflow automation. The fastest way to reach me is email.