Sohag Gain
Multi-agent orchestration
Separate research, sales, CRM, communication, and reporting responsibilities so each decision path can be inspected and improved.
LangGraphHuman approvalREST + MCP
Inspect architecture →
AI Automation Engineer and AI Agent Developer who has delivered 100+ projects for 25+ clients, designing and shipping production agents, RAG pipelines, API integrations, and CRM workflows. The work also draws on AI Engineering and AI Solutions Architecture, using Python, FastAPI, n8n, Make.com, Zapier, and GoHighLevel.
These versions keep the same experience, projects, education, and research while tailoring the opening summary to the role.
Design and deploy AI-driven automation systems — CRM workflows, lead-routing pipelines, and AI agents — using GoHighLevel, n8n, Make.com, and Zapier, integrated with Claude, OpenAI, and Gemini APIs. Selected systems are published as reviewable repositories.
Advised clients on automation strategy and CRM optimization, connecting Facebook Lead Ads, HubSpot, Gmail, and Slack into unified pipelines.
Delivered web design and development for international clients while completing my BSc — the API-integration foundation behind my current automation work.
Selected public builds show the system boundaries, approval points, and interfaces behind the role title.
Separate research, sales, CRM, communication, and reporting responsibilities so each decision path can be inspected and improved.
Document ingestion, chunking, embedding, retrieval, and answer generation are designed as a traceable retrieval pipeline.
A custom MCP server centralizes authenticated access to CRM, database, email, calendar, search, and analytics tools.
Large-scale Bangla–English text-classification research informs how I evaluate model quality, baselines, and deployment trade-offs.
My BSc thesis evaluates five classical models, BiLSTM, BERT, and LSTM+BERT on 562,163 news articles using a unified 80/10/10 evaluation setup. In the reported test results, BERT reached 92.95% accuracy in Bangla and 94.50% in English; TF-IDF + SVM exceeded 90% on both datasets as a lower-compute baseline. The thesis also reports that 98.8% of the Bangla articles fall in the medium-to-high inflectional-density bands.
Proven in production + working daily. Currently learning lives only on the home page.
LEDP · 2020–2021
Remote, hybrid, full-time, part-time, or contract — I'm listening.
