# Masum Kabir Biswas > Masum Kabir Biswas is an AI/ML Engineer in Kolkata, India. I build AI agents and ML systems that run in production, plus the products around them. - Website: https://masumkabirbiswas.com - LinkedIn: https://www.linkedin.com/in/masum-kabir-biswas-17058726a - GitHub: https://github.com/KKABIR07 - Email: kabirbiswas131@gmail.com - Sitemap: https://masumkabirbiswas.com/sitemap.xml ## Notes - [Anatomy of a real-time AI voice agent](https://masumkabirbiswas.com/notes/anatomy-of-a-voice-agent): What it takes for an AI to hold a phone call: voice activity detection, streaming speech-to-text, a grounded LLM, fast speech synthesis, and a clean hand-off to a human. - [Askora: an AI workstation that runs from a USB drive](https://masumkabirbiswas.com/notes/askora-offline-ai-workstation): Chat, web search, image and video generation, an in-browser video editor, a medical report reader and a 3D anatomy atlas, all running offline from one portable folder on a 4 GB laptop GPU. - [How this portfolio works](https://masumkabirbiswas.com/notes/how-this-portfolio-works): A talking intro video, a 3D mini-me you can ask questions out loud, narrated demo videos, and hosting that uses about 70 MB of RAM. Here is what is behind each piece. - [Running an AI agent team from Telegram](https://masumkabirbiswas.com/notes/hermes-openclaw-agent-team): How Hermes and OpenClaw split projects, ship code under their own GitHub accounts and run marketing for three brands, with a human approving everything public. ## Who Masum is Masum Kabir Biswas is an AI/ML Engineer based in Kolkata, India. I build AI agents and ML systems that run in production, plus the products around them. Technical Team Lead at Neuraforz. I design multi-agent systems, voice and speech pipelines and LLM tooling, then ship them with the full-stack apps that let people steer them. ## About Masum I started in Information Technology at MAKAUT and spent the last year putting language models to work: agents that write, publish, audit sites, find clients, render video and hand tasks to each other over Telegram. ## About Masum Most of what I build is the plumbing nobody sees. That means queues, schedulers, approval gates, tenant isolation and deploy scripts, so an AI system can run unattended without doing something embarrassing. ## About Masum I'm also studying Biochemistry at Maulana Azad College, because I'm curious where AI meets the life sciences. ## How to contact or hire Masum To work with Masum, use the contact form at the bottom of the site, email kabirbiswas131@gmail.com, message him on LinkedIn (https://www.linkedin.com/in/masum-kabir-biswas-17058726a), or chat with him on WhatsApp (link in the footer). His code is on GitHub (https://github.com/KKABIR07) and his photography is on Instagram (https://www.instagram.com/shutter.poetry.1). He is open to AI/ML engineering roles, freelance AI agent and automation projects, and full-stack product work. He usually replies within a day. Time zone: India (IST). ## Projects overview: everything Masum has built Masum has built 6 featured projects, all in production: Hermes & OpenClaw (Autonomous agent team · Neuraforz); Zap Voice Agent (Real-time AI calling agent); Team Tracker (Desktop app + browser extension · Neuraforz); EventOS 2.0 (Event & livestream platform); NeuraLeads (AI lead-gen SaaS · Exzelon); AI Video Stack (Generative video studio as MCP tools). Each has a narrated demo video and a system diagram in the Selected work section. ## Project: Hermes & OpenClaw Hermes & OpenClaw (Autonomous agent team · Neuraforz). Two self-hosted AI agents on Oracle Cloud that run marketing, SEO, video and lead generation for three brands. You direct them from Telegram. Tag Hermes with a project and he splits the work, hands half to OpenClaw, and both commit to GitHub under their own bot accounts. The X agent posts news carousels and videos for Neuraforz, Exzelon and Medeoan through an approval queue with daily caps. The SEO autopilot runs 18 audit playbooks every three days, fixes what it can and pushes a branch for review. The client finder pulls US/EU leads from job boards every day and sends a CSV to Telegram. Tech stack: Python, DeepSeek, Playwright, systemd, MCP, HyperFrames, Tailscale. Key numbers: 2 agents, 3 brands, 20+ scheduled jobs. Live at https://x.com/neuraforz. ## Project: Zap Voice Agent Zap Voice Agent (Real-time AI calling agent). An AI voice agent that makes B2B discovery calls. It listens, answers from a knowledge base, scores the lead and hands hot prospects to a human. Streaming voice loop built on Pipecat: voice-activity detection, Whisper speech-to-text on Groq, Llama 3.1 for reasoning, Kokoro for speech. Answers come from a RAG knowledge base, so the agent stays on the facts of the product it's selling. Every call ends with a lead score; hot leads trigger an n8n workflow that alerts a sales rep. Tech stack: Python, Pipecat, Whisper, Llama 3.1, Kokoro TTS, RAG, n8n, MongoDB. Key numbers: Voice real-time loop, RAG grounded answers, Live in production. Live at https://www.neuraconnects.xyz/. ## Project: Team Tracker Team Tracker (Desktop app + browser extension · Neuraforz). A privacy-first work-time tracker in the spirit of Hubstaff. A Windows desktop app and a Chrome extension record whether people are working, never what they type, and managers review it in a web dashboard. Desktop app built with Tauri and Rust: tray icon, consent screen, live tracking indicator, a break button, and in-app auto-updates shipped through GitHub Releases. A Rust agent samples activity and app usage, buffers it in a local SQLite queue so nothing is lost offline, and uploads in batches. A Manifest V3 Chrome extension logs time per website domain only, never full URLs or page content, and hands it to the desktop app over a localhost bridge. NestJS and Postgres backend with timesheets, activity charts and a screenshot gallery, plus a machine-to-machine API that feeds Exzelon’s performance platform. Tech stack: Tauri 2, Rust, SQLite, Chrome MV3, NestJS, Prisma, PostgreSQL, Next.js. Key numbers: Desktop Tauri + Rust, Chrome MV3 extension, v1.1.7 auto-updating. Live at https://neuratracker.neuraforz.com/download. ## Project: EventOS 2.0 EventOS 2.0 (Event & livestream platform). A SaaS for running events end to end: ticketing, a marketplace, and live streaming with chat, polls and Q&A. WebRTC streaming through a mediasoup SFU, with Socket.IO chat scaled across instances by a Redis adapter. QR tickets issued at checkout and scanned at the door, plus an analytics dashboard with Excel export. Multi-tenant organisations, role-based access and JWT refresh-token rotation. Tech stack: Next.js 16, React 19, Express 5, mediasoup, Socket.IO, MongoDB, Redis. Key numbers: WebRTC live video, SFU fan-out, CI/CD GitHub Actions. Live at https://eventosonline.xyz. ## Project: NeuraLeads NeuraLeads (AI lead-gen SaaS · Exzelon). A multi-tenant platform that finds hiring companies, enriches the contacts and runs cold-email campaigns with an AI inbox. Pulls leads from job sources, then enriches people and company data through pluggable adapters. Runs campaigns across about 500 warmed mailboxes, with AI reply suggestions and spam checks. Its MCP connector exposes 148 tools, so assistants like Claude can operate the platform directly. Tech stack: FastAPI, SQLAlchemy, MySQL, Redis, Next.js, MCP. Key numbers: 489 API routes, 148 MCP tools, ~500 mailboxes. Live at https://neuraleads.ai. ## Project: AI Video Stack AI Video Stack (Generative video studio as MCP tools). A self-hosted video studio on a NAS that AI agents call as tools. Ask for a video and you get back a narrated, scored, finished MP4. About 30 MCP tools covering script-to-speech (Kokoro, voice cloning), Whisper word timing, music, mixing and loudness checks. Renders with Remotion and HyperFrames pipelines, and cuts long videos into shorts automatically. Per-user API keys, isolated job queues and outputs, so a whole team and the Hermes video agent share one box. Tech stack: Python, MCP, Whisper, Kokoro TTS, Remotion, FFmpeg, Docker. Key numbers: ~30 MCP tools, 45 s 1080p render, 13 video pipelines. ## Experience: Technical Team Lead at Neuraforz Technical Team Lead at Neuraforz (Full-time), Oct 2026 to Present, Kolkata, West Bengal, On-site. I lead the engineering team and own the architecture of the agent platform and client products. Skills: AI, Software architecture, Team leadership. ## Experience: Software Engineer, AI/ML at Neuraforz Software Engineer, AI/ML at Neuraforz (Full-time), Feb 2026 to Sep 2026, Chicago, Illinois, Remote. I built LLM agents and Next.js/FastAPI apps, and deployed 10+ live services on VPS infrastructure. Skills: Machine learning, LLM agents, FastAPI. ## Experience: Application Developer at Exzelon Application Developer at Exzelon (Part-time), May 2026 to Present, Hyderabad, Telangana, Remote. I build and ship the recruiting and lead-generation products on Next.js and Node. Skills: Node.js, Next.js. ## Experience: Junior Software Engineer at Infomaticae Junior Software Engineer at Infomaticae (Full-time), Aug 2025 to Nov 2025, Kolkata, West Bengal, Hybrid. I shipped features across Python services and React front-ends. Skills: Python, React.js. ## Career progression Masum joined Neuraforz as Software Engineer, AI/ML in Feb 2026 and was promoted to Technical Team Lead in Oct 2026. He went from Junior Software Engineer (Infomaticae, Aug 2025) to team lead in about fourteen months. ## Skills AI / ML: LLM agents, Multi-agent systems, RAG, Speech (Whisper, TTS), Machine learning, MCP servers, Claude API, n8n. Front-end: Next.js, React, TypeScript, Tailwind CSS, Motion. Back-end: Python, FastAPI, Node.js, Express, Prisma, SQLAlchemy. Data: PostgreSQL, MongoDB, MySQL, Redis. Infrastructure: Docker, Nginx, PM2, AWS, CI/CD, Linux. ## Education Where Masum studied: B.Tech, Information Technology at Maulana Abul Kalam Azad University of Technology, 2021 – 2025 (Grade 8.5). B.Sc, Biochemistry at Maulana Azad College, 2026 – 2028 (In progress). He is from Kolkata, West Bengal, India. ## Certifications Masum holds 10 certifications: Machine Learning and Image Processing (Machine learning; Cisco, Dec 2021); Master Amazon EC2 Storage: Complete Guide for EBS, EFS & AMI (AWS; Udemy, Nov 2025); Next.js 15 & React: The Complete Guide (Next.js; Udemy, Nov 2025); SQL for Beginners: Learn SQL using MySQL and Database Design (SQL; Scaler, Jan 2025); Node JS Certification Course: Master the Fundamentals (Node.js; Scaler, Jun 2024); HTML and CSS in depth (HTML & CSS; Coursera, Jun 2024); React Basics (React; Coursera, May 2024); Programming with JavaScript (JavaScript; Coursera, May 2024); The Complete 2024 Web Development Bootcamp (Web development; Udemy, Apr 2024); Unreal Engine 5 C++ Developer: Learn C++ & Make Video Games (C++; Udemy, Feb 2024). Each has a verification link in the Certifications section of the site. ## About this website and the mascot This portfolio is a Next.js site with motion design. The small character is mini-Masum, a 3D version of Masum that walks around the page and answers questions about him. It only knows what is on this site.