I design and build AI agents and dependable automation systems—from workflow architecture and integrations to deployment. My work also covers AI engineering and AI solution architecture.
Portfolio totals include private and client-specific work. 17 public GitHub repositories are linked in the Portfolio.
Four areas where automation moves the needle most.
Agents help with email, calendar, support, and lead qualification using Claude, OpenAI, or Gemini, with memory and connected tools.
n8n, Make, and Zapier connect CRMs, inboxes, and marketing tools so repetitive steps happen automatically.
CRM pipelines, snapshots, white-label builds, and AI assistants configured around the tools your team already uses.
Documents become a searchable knowledge base, allowing an AI assistant to answer from your own information.
Simplified simulation of my Gmail & Calendar Assistant — one goal in, the agent plans and executes.
"Find tomorrow's available meeting slots and email John."
Email sent to John — 10:00–10:30 AM proposed, calendar invite attached.
An example lead-qualification flow. Choose a detail level, then select a step to explore it.
Someone sends a form, message, or call request. That starts the follow-up process.
Simplified view of my RAG pipeline — documents in, grounded answers out.
"What's your refund policy?"
"Refunds are available within 30 days of purchase, provided the item is unused and in its original packaging. Contact support with your order number to start a return."
One pipeline shape — a request comes in, an AI agent reasons, automation executes, the result lands where your team can see it.
Something happens — a form fill, a message, a missed call.
A form submission, inbound message, missed call, or scheduled trigger fires a webhook, passing a structured payload into the pipeline.
AI reads the request and figures out what to do next.
Claude, GPT-5, or Gemini processes the payload — using function calling and structured outputs to classify, qualify, or decide the next action.
The right steps happen automatically, in the right order.
n8n, Make.com, or Zapier orchestrates branching logic, retries, and routing across connected nodes.
The AI talks to the other apps your business already uses.
REST APIs and webhooks pass authenticated requests between the AI layer and connected tools.
The record gets saved and moved to the right stage.
GoHighLevel or HubSpot's API updates the contact record, sets the pipeline stage, and triggers the next touchpoint.
Information is stored so it can be found again later.
PostgreSQL or a vector store (Pinecone) persists structured data or embeddings for retrieval.
The task gets done — no one had to do it by hand.
The pipeline resolves to a logged, auditable outcome — timestamped, with a status a human can check.
The same lead, handled two ways.
Two things separate a real system from a demo — when to ask a human, and what to do when something breaks.
For anything with real stakes — money, a customer message, a record change — the system pauses for a human instead of guessing.
Good automation doesn't only handle the happy path. It knows what to do — and who to tell — when something goes wrong.
Grouped by what they do in a real system. Hover for detail.
Proven in production, working daily, or actively learning — filter to explore.
LEDP · 2020–2021
Foundation program — the API-integration base behind my automation work.
BSc in Computer Science & Engineering. I design the pipelines that qualify leads, answer customers, and close the loop — so teams stop copying and start shipping.
Dhaka · Remote (GMT+6) · sohag@sohaggain.com
Selected systems and AI agents, with workflows and source to explore.
Multi-agent assistant for email and calendar from text or voice, with contact lookup and memory.
Documents into Pinecone as embeddings, then grounded answers with memory-enabled semantic search.
Answers service questions, qualifies leads, detects booking intent — with human handoff for edge cases.
Not adjectives — here's what you're getting.
Start with the bottleneck, not the tool — so the system fixes what's slowing you down.
BSc in CSE — APIs, data, and logic under the workflow, not just drag-and-drop.
AI for judgment calls, deterministic automation for everything else.
Discovery → design → build → test → handover. One owner, no handoff gaps.
Before building production AI systems, I tested how model choices perform across languages, data scales, and resource constraints.
My BSc thesis evaluates classical machine learning, BiLSTM, BERT, and LSTM+BERT across 562,163 news articles. In the reported evaluation, BERT reached 92.95% test accuracy for Bangla and 94.50% for English; TF-IDF + SVM provided a lower-compute baseline above 90% for both datasets.
Selected work shown with client permission. Project links show the delivered system; outcome language is kept to the original client-provided review unless independently documented.
Data Scraping & Offer Monitoring
Evidence: client-provided testimonial · system preview linked
Evidence: client-provided testimonial · system preview linked
Reusable Google Maps Local Business Lead Generator
Evidence: client-provided testimonial · system preview linked
Gulf Bangla News — AI WordPress to Facebook Automated Publisher
Evidence: client-provided testimonial · system preview linked
Brother Sohag, you did an amazing job! I am so happy with the final outcome. Your dedication and quality of work are truly outstanding.
— Shovon · Data Scraping & Offer Monitoring
See related project →Incredible work by Sohag bhai! Our Messenger automation is running flawlessly—instant replies, clear conversation routing, and zero bugs. The whole workflow feels natural and professional. I am extremely impressed with the quality and turnaround time. Highly recommended for any automation project!
— Showan Howlader · Messenger Automation
See related project →Pure precision and efficiency. Sohag bhai doesn't just build WhatsApp bots; he builds scalable business systems. The logic trees, API triggers, and fail-safes he implemented cut our team's manual workload by over 70% from day one. Flawless architecture, zero downtime, and a masterclass in modern automation. Worth every single penny.
— Mustaque Ahmed · WhatsApp Automation
The workload reduction figure is client-reported; a measurement method or independent record is not linked here.
See related project →I was blown away by the AI calling agent Sohag built for us. Most voice bots sound rigid and robotic, but this one carries conversations with natural pacing, handles subtle interruptions smoothly, and sounds remarkably human. It immediately took over our inbound qualification and calendar scheduling without missing a beat. Exceptional engineering and seamless execution.
— Asif Iqbal · AI Calling Agent Automation
See related project →Working with Sohag was a fantastic experience. He delivered top-notch work well ahead of the deadline, communicated clearly throughout, and paid great attention to every detail. Truly exceeded expectations!
— Carine · Reusable Google Maps Local Business Lead Generator
See related project →Incredible service and great attention to detail. Sohag understood the requirements right away and executed everything flawlessly. Fast, reliable, and highly professional. Will definitely work with him again.
— Ibrahim Hossain · Gulf Bangla News — AI WordPress to Facebook Automated Publisher
See related project →Remote, hybrid, freelance — let's figure out what actually needs automating.
Sohag Gain
