Suvon Roy — AI Systems Engineer & Automation Architect

01 — About

The Engineer
Behind the Systems

📍 Dhaka, Bangladesh

Available globally · Remote-first

I'm Suvon Roy — a Full-Stack Python Developer and AI Systems Engineer who builds things that actually work in production. Not prototypes. Not demos that crash on Tuesday. Systems that handle edge cases, recover from failures, and deliver value every single day.

My work spans AI agent pipelines, voice systems, WebRTC communications, and intelligent automation. My most technically challenging project — a headless WebRTC intercom client running on Raspberry Pi for a German enterprise client using Riedel's STAGE™ platform — ships Opus audio through a full Keycloak → SockJS → WebRTC stack with zero manual intervention on embedded Linux.

I operate in the space where AI meets infrastructure. I use Groq, Claude, Vapi, and n8n as tools — not magic — to solve problems that have real cost implications. When I automate a workflow, I measure the hours saved. When I build an AI agent, I track conversion rates.

Currently a CSE student at Dhaka International University — but the degree is a formality. I've been shipping production code for clients in India, USA, UAE, UK, Canada, and Germany since before most developers I know started their courses.

Primary Stack
Python · FastAPI · React

AI / Voice
Groq · Vapi · XTTS v2

Automation
n8n · Make · Playwright

Availability
Open for Remote Contract

Start a Project ↓ Download CV

02 — Skills

Technical
Expertise

AI & LLM Engineering

Groq API Claude API OpenAI Vapi Voice AI XTTS v2 / Coqui RAG Pipelines Prompt Engineering LangChain Whisper STT

Backend & APIs

Python FastAPI Django Node.js WebRTC / SFU WebSockets / SockJS REST API Design PostgreSQL Redis

Automation & Workflow

n8n Make.com Playwright Selenium Puppeteer Scrapy Webhook Pipelines CAPTCHA Solving

Frontend & UI

React Next.js 14 Vite TypeScript Tailwind CSS HTML / CSS / JS

Infrastructure & Embedded

Docker Linux / systemd Raspberry Pi / ARM64 PulseAudio / ALSA Keycloak OIDC Railway · Render · Vercel Supabase PM2

03 — Services

What I
Build

Specialized AI and automation engineering for businesses that need systems — not experiments. Every engagement starts with a clear outcome and ends with measurable results.

01

AI Agent & Voice Pipeline Development

Production AI agents for inbound/outbound calls, customer support automation, lead qualification, and conversational interfaces — integrated with your existing CRM and tools.

VapiGroqClaude APIFastAPI

→

02

n8n / Make Workflow Automation

End-to-end business process automation — multi-step CRM workflows, AI content pipelines, lead intake systems, email sequences, and multi-platform integrations that run without oversight.

n8nMake.comWebhooksAI / LLM

→

03

Full-Stack SaaS Development

Multi-tenant SaaS platforms with authentication, subscription billing, real-time features, and AI integration. From MVP to production-ready in weeks, not months.

FastAPINext.js 14SupabaseStripe

→

04

WebRTC & Real-Time Communications

Full-duplex audio/video systems, headless browser clients for embedded hardware, intercom integrations, and peer-to-peer communication infrastructure built for reliability.

WebRTCNode.jsLinuxEmbedded

→

05

Intelligent Web Scraping & Data Pipelines

Scalable data extraction from complex, dynamic sites with session management, AI-assisted classification, deduplication, and structured delivery to your database or dashboard.

PlaywrightScrapyPythonData Pipeline

→

06

AI-Powered Desktop Applications

System-level AI tools with voice control, screen awareness, multi-provider LLM support, and native OS integration — packaged for Windows, Linux, or embedded deployments.

PythonPyInstallerVoice AISystem Control

→

04 — Work

Featured
Projects

001 Client · Germany Deployed

Riedel STAGE™ Headless Intercom Client

A German broadcast client needed a Riedel intercom node without buying a physical SmartPanel (~€2k). I reverse-engineered the STAGE™ WebSocket protocol from browser traffic and built a Node.js client that runs the full intercom stack headlessly on a Raspberry Pi 4 — Keycloak auth, SockJS signaling, WebRTC audio, PulseAudio FIFO pipeline, all packaged as a systemd service.

Node.js WebRTC / Opus Raspberry Pi ARM64 Keycloak OIDC PulseAudio SockJS / STOMP systemd

Context

The Riedel STAGE™ Virtual SmartPanel is a browser-only intercom interface — it only runs inside Chromium. The client needed it deployed on embedded ARM64 hardware with no monitor, no GUI, and real microphone I/O from a HiFiBerry ADC Pro hat.

What I Built

Captured and decoded Riedel's proprietary SockJS/STOMP wire format from browser DevTools — no SDK or docs available
Full Keycloak OIDC token refresh loop — sessions stay alive for 8+ hour shifts without re-auth
arecord → named FIFO → Chromium fake device pipeline; solved SIGPIPE watchdog crash that killed audio on reload
Single-command Pi installer, ARM64 native addon build, systemd unit with restart policy

Outcome

Runs fully headless on a Pi 4 — no display, no keyboard, boots straight into intercom duty
Replaced a €2,000 hardware unit; client used it in a live production facility
Stable across multi-hour sessions — no dropouts observed during production runs
Hardest part: the FIFO/SIGPIPE bug that only appeared after a watchdog-triggered Chromium reload

↗ Live Demo GitHub

002 Deployed

Signature Roof Co — AI Voice + Chat Demo

A working demo I built for roofing SMBs — not a pitch deck, an actual product. A company site, a Groq-powered chat widget, an inbound Vapi voice agent that qualifies callers and books appointments, and a FastAPI backend logging every lead. Used it directly to close automation contracts on Fiverr and Upwork.

Vapi Groq FastAPI Voice Agent Lead Capture Railway

Why I Built It

Most SMB owners have heard "AI agent" but never seen one working on their phone line. I wanted something I could show a roofing owner directly — here's your missed call answered, here's the lead logged. Built it in a week to use as a concrete sales tool.

How It Works

"Jordan" — a Vapi voice agent that picks up inbound calls, runs through qualifying questions, and schedules a callback
Groq/LLaMA chat widget embedded on the site, system-prompted for roofing-specific Q&A
FastAPI backend writes every conversation turn and lead contact to Supabase
Single Railway deployment — backend + frontend on one domain, one command

What Came of It

Closed 3 automation contracts by linking clients directly to this demo during discovery calls
Voice qualification averaged ~75 seconds per caller vs 4–6 minutes on a human call
Groq free tier kept inference cost at zero during all demos; total infra cost was Vapi per-minute billing only
Proved a working AI stack for a small business is days of work, not months

↗ Live Demo GitHub

003 Product — Gumroad Live Demo

NEXUS AI — Local Windows AI Assistant

A JARVIS-style desktop assistant I built and sold as a product on Gumroad at $29/license. Wake-word detection, clipboard + screen awareness, process monitoring, multi-provider LLM support (Groq, Gemini, Claude, Ollama), and a holographic Tkinter GUI. Packaged as a .exe with a FastAPI license validation backend — so I could actually sell and revoke keys.

Python Groq / Gemini / Claude Wake Word Tkinter GUI PyInstaller License System

What I Was Solving

Every AI tool lives in a browser tab. I wanted one that knows what process is eating my CPU, can read what I just copied, see what's on my screen, and respond when I say its name — no tab switching, no typing. And I wanted to sell it, not just build it.

Technical Decisions

Wake-word runs entirely local via Porcupine — no audio sent to cloud, activates in <100ms
LLM provider is a swappable config — Groq for speed, Ollama for offline, Claude/Gemini for quality
Screenshot and clipboard context injected into every prompt automatically — the AI sees what you see
License keys validated against a FastAPI server I host — real revocation, not honor-system DRM
Compiled to .exe via PyInstaller; no Python install needed on the buyer's machine

Shipped & Sold

Listed on Gumroad at $29/license — actual paying customers, not just downloads
Voice response latency under 200ms on Groq; acceptable on Ollama with local models
First project I went through the full loop on: build → package → license server → sell → support
Learned more about distribution than development building this one

↗ Live Demo ↗ Gumroad — $29

004 SaaS Platform Live Demo

VoiceForge AI — TTS SaaS with XTTS v2

Built a full ElevenLabs alternative using Coqui XTTS v2 — multi-speaker TTS with voice cloning, real-time audio streaming, three-tier billing (Free / Pro / Enterprise), Supabase auth, Upstash rate limiting, and a Next.js 14 frontend. Entire stack runs at ~$0/month on free tiers. Shipped in about 12 days.

FastAPI XTTS v2 Next.js 14 Supabase Upstash Redis Stripe

Starting Point

ElevenLabs is $22+/month. XTTS v2 is open-source and produces similar quality. I wanted to see if I could wrap it in a real SaaS — with auth, billing tiers, rate limiting, and a proper frontend — and have it cost me nothing to run while I tested it.

Stack Choices

FastAPI streaming endpoint — XTTS v2 generates chunks, client receives audio as it renders, no waiting for full file
Supabase Auth with JWT; role stored in user metadata, checked on every API request
Upstash Redis for per-user rate limiting — sliding window, 50 req/day free, configurable per tier
Next.js 14 with RSC — most pages are server-rendered, only the audio player is a client component
Railway for the FastAPI backend, Vercel for the frontend, Supabase for DB — all free tier

Honest Assessment

Shipped in ~12 days; the auth + billing wiring took longer than the XTTS integration
Voice quality is close to ElevenLabs on clean input; noticeably worse on fast or accented speech
Rate limiting actually mattered — without it, a single user could exhaust Railway's free CPU in minutes
The zero-cost stack works for demos and low traffic; XTTS is too slow for high-volume production

↗ Live Demo GitHub

005 Client Work Live Demo

KHANS Codes & Standards — Lead & CRM Automation

Built the lead intake and follow-up automation for a real client (KHANS Codes & Standards) using n8n and Make.com. Webhook from their form triggers qualification, CRM logging, and a 5-step SMS + email sequence — all firing within seconds of submission, no manual work from the sales team.

n8n Make.com CRM SMS Automation Email Sequences Webhooks

Situation

KHANS's sales person was copy-pasting form submissions into a spreadsheet, then sending follow-up emails one by one. Leads submitted on weekends or evenings often didn't get a reply until the next business day. They asked me to fix it without changing how their form worked.

What I Built

n8n webhook catches the form POST, parses the lead, and writes it to their CRM in the right pipeline stage
First SMS fires within ~15 seconds; email sequence continues over 3 days with conditional branching
Make.com scenario sends a Slack/email alert to the sales person with the full lead context
Did not touch their existing form or website — webhook only, no code deployed to their server

Result

First response now goes out in under 30 seconds regardless of time of day
Their sales person stopped doing manual data entry — that was roughly 8–10 hours/week recovered
Follow-up rate went from whatever they remembered to do, to every lead getting the full sequence
Client runs it unattended; I haven't touched it since delivery

↗ Live Demo GitHub

05 — Experience

Where I've
Worked

AI SaaS Builder

Freelance Engineer

KHANS C&S

Education

Independent AI SaaS Developer

Self-Directed · Remote 2023 – Present

Built and shipped 3 production SaaS platforms: CallSaaS (multi-tenant AI calling), VoiceForge AI (ElevenLabs-alternative TTS), and AI Workforce Platform (virtual employee system) — each with full auth, billing, and rate limiting.
Engineered a zero-cost SaaS architecture using Groq (free-tier LLM inference), Supabase (auth + database), Railway (backend hosting), and Vercel (frontend) — reducing operational cost to near $0.
Shipped NEXUS AI, a monetized J.A.R.V.I.S.-style Windows assistant sold on Gumroad at $29/license with a built-in license validation system.
Deployed production AI agents on GratisVPS (6GB RAM, no card) using PM2, Groq API, and Supabase — proven zero-cost hosting stack for AI workloads.

FastAPINext.js 14GroqXTTS v2SupabaseReact/VitePM2

Senior Freelance Python & Automation Engineer

Fiverr · Upwork · Direct Clients 2022 – Present Remote · Worldwide

Delivered 15+ automation and full-stack projects for clients in India, USA, UAE, UK, Canada, and Germany across two years of active contracting.
Built the Riedel STAGE™ headless intercom client for an enterprise broadcast facility in Germany — the most technically complex WebRTC/embedded Linux project in the client's vendor history.
Engineered web scraping frameworks for JS-heavy and protected platforms with rotating residential/mobile proxies and session management for ongoing client data operations.
Built n8n and Make.com workflows for AI content pipelines, YouTube channel management, and email enrichment systems for business clients globally.
Maintained 5-star ratings across platforms through clean code, thorough documentation, and proactive post-delivery support.

PythonSeleniumPlaywrightScrapyn8nMake.comFastAPI

Automation Engineer — CRM & Lead Systems

KHANS Codes & Standards 2025 – Present Remote · Client Project

Architected an end-to-end CRM lead intake and follow-up automation system using n8n and Make.com as the backbone — from form submission to qualified lead in under 30 seconds.
Designed webhook-driven pipelines that route, qualify, and log leads into the CRM in real time, eliminating manual data entry entirely.
Built 5-step personalized follow-up sequences delivered via SMS and email with dynamic lead data insertion and configurable timing.
Coordinated system access, OAuth integrations, and live deployment with a cross-functional team under a tight delivery timeline.

n8nMake.comCRMWebhooksSMS AutomationEmail Sequences

B.Sc. in Computer Science & Engineering

Dhaka International University 2022 – Present Dhaka, Bangladesh

Pursuing a full CSE degree while simultaneously building and shipping production systems for paying clients — the two tracks directly inform each other.
Self-taught across Python automation, full-stack web development, AI/ML integration, WebRTC, embedded Linux, and cloud infrastructure in parallel with formal coursework.
Strong foundation in algorithms, data structures, and software architecture that directly informs the design of every production system built professionally.

CSEAlgorithmsSoftware ArchitectureSelf-Taught AI/ML

06 — Reviews

What Clients
Say

★★★★★

The AI workflow he built for our lead system was a game changer. Every lead now gets followed up within 30 seconds — automatically. Our team's time is freed up for actual closing.

A

Ahmad K.

Operations Director

Direct

★★★★★

Built an n8n workflow that saved our marketing team 20+ hours a week. Communication was extremely clear throughout. Delivery was ahead of schedule with excellent documentation.

S

Sarah K.

Marketing Manager, USA

Upwork

★★★★★

The web scraping framework extracts data from sites I thought were completely locked down. Clean code, great error handling, and he explained every technical decision. Will hire again.

T

Thomas B.

Data Analyst, UK

Fiverr

★★★★★

Rare combination: technically exceptional AND a great communicator. He understood our requirements faster than any developer I've worked with and delivered a system that's been running flawlessly for months.

M

Michael R.

Founder, UAE

Direct

★★★★★

Our AI video pipeline in n8n goes from a brief text prompt to a finished video in under 10 minutes. It replaced two freelancers and an entire afternoon of manual work every single day.

L

Lisa W.

Content Director, Canada

Upwork

★★★★★

He solved a WebRTC problem in 48 hours that our in-house team had been struggling with for weeks. Deep technical knowledge, zero hand-holding needed, and genuinely proactive about edge cases.

J

Jonas H.

CTO, Germany

Direct

★★★★★

The AI workflow he built for our lead system was a game changer. Every lead now gets followed up within 30 seconds — automatically. Our team's time is freed up for actual closing.

A

Ahmad K.

Operations Director

Direct

★★★★★

Built an n8n workflow that saved our marketing team 20+ hours a week. Communication was extremely clear throughout. Delivery was ahead of schedule with excellent documentation.

S

Sarah K.

Marketing Manager, USA

Upwork

07 — Writing

Technical
Articles

WebRTC · Embedded

Building a Headless WebRTC Intercom on Raspberry Pi: The Complete Guide

How I reverse-engineered a proprietary intercom protocol, solved Keycloak token refresh edge cases, and piped real microphone audio through PulseAudio FIFOs into a headless Chromium instance.

2025-04 Read Article →

SaaS · Infrastructure

The Zero-Cost AI SaaS Stack: Groq + Supabase + Railway + Vercel

A practical breakdown of how to build and deploy a production multi-tenant AI SaaS with user auth, rate limiting, and payment integration — running at $0/month on free-tier services until you hit real scale.

2025-03 Read Article →

n8n · Automation

n8n vs Make.com in 2025: A Production Developer's Honest Comparison

After building over 20 workflows across both platforms for real clients, here's what I've learned about when to use each one — including the edge cases, pricing gotchas, and hidden limitations nobody talks about.

2025-02 Read Article →

08 — Contact

Let's Build
Something

Ready to talk about your project?

I'm available for remote contracts, long-term engagements, and consulting. Whether you need an AI agent pipeline, a SaaS MVP, or an automation system — I'll give you a direct answer about what's buildable and what it takes.

Email

piyashb55@gmail.com

+880 1604 430796

linkedin.com/in/suvon-roy-a2a7203a6

Available for contracts — Responding within 12 hours