Building AI that operates autonomously

Anuj · M4ST · Bareilly, India · Est. 2024

AI Engineer
& Orchestrator

Multi-Agent Systems · MCP Protocol · Zero Cost

I build autonomous agent systems that genuinely think independently. 21 MCP servers, 11 free LLM providers, ChromaDB semantic memory — all running locally at zero recurring cost. Production-grade AI without the cloud bill.

View Projects Main Portfolio Hire Me
0 MCP Servers
0 LLM Providers
$0 Monthly Cost
∞ Scale
// Autonomous by design · Local-first architecture

ABOUT

Hi, I'm Anuj
aka M4ST

I'm a solo AI systems engineer from Bareilly, Uttar Pradesh, India. I specialize in building autonomous multi-agent systems — AI pipelines that decompose tasks, collaborate across specialist agents, maintain semantic memory, and execute without human hand-holding.

My entire stack is local-first — 21 MCP servers, 11 free LLM providers routed intelligently via LiteLLM, ChromaDB for persistent vector memory, and CrewAI/LangGraph for orchestration. The entire system costs exactly $0 per month to run in production.

I'm available for AI freelance engagements worldwide — agent system design, MCP server development, LLM pipeline automation, and AI infrastructure consulting.

01
Local-First
Everything runs on your hardware. No cloud dependency, no data leakage, no monthly SaaS bills.
02
Zero Cost
Production AI at $0/month. Free LLMs routed intelligently. Enterprise capability without enterprise cost.
03
True Autonomy
Agents that plan, collaborate, remember, and execute. Not wrappers — genuine autonomous systems.
04
Composable
21 MCP servers. Mix and match. Extend infinitely. Every component is a building block.
m4st_agent.py
1from crewai import Crew, Agent, Task
2from langchain_community.vectorstores import Chroma
3
4# 21 MCP servers, 11 free LLM providers
5orchestrator = Agent(
6 role="Master Orchestrator",
7 goal="Decompose & delegate autonomously",
8 llm="gemini/gemini-1.5-pro", # free
9 memory=True,
10 tools=mcp_tools[:21]
11)
12
13# $0/month. Fully autonomous. 🔥
14result = Crew(agents=[orchestrator]).kickoff()
stack.json
1{
2 "mcp_servers": 21,
3 "llm_providers": 11,
4 "monthly_cost": 0,
5 "architecture": "local-first",
6 "memory": "ChromaDB + semantic",
7 "autonomy": "genuine",
8 "status": "production"
9}

PROJECTS

What I've built

⚡
↗

FLAGSHIP PROJECT

M4ST Agent Orchestration Stack

The flagship M4ST system — a local-first, multi-agent orchestration platform built on LangGraph and CrewAI. Coordinates specialized agents across 21 MCP servers with ChromaDB-powered persistent semantic memory, intelligent LLM routing across 11 free providers, autonomous task decomposition, and true zero-cost operation. This runs in production, offline, on consumer hardware.

LangGraphCrewAIChromaDB MCP ProtocolPythonOllama LiteLLMFastAPI
Production · Live
Local-first · $0/month
🔍
↗

Autonomous OSINT System

Self-directed intelligence gathering with multi-source fusion, anomaly detection, pattern recognition across temporal data, and automated report generation. Zero human intervention once tasked.

LangChainAgentsOSINT APIs
Live
Autonomous
🧠
↗

Semantic Memory Engine

Persistent vector memory with intelligent retrieval. Semantic clustering, relevance decay, context-aware injection, and cross-session memory consolidation for agents that genuinely remember.

ChromaDBEmbeddingsRAG
Production
Vector Memory
🐙
↗

Smart GitHub Manager

AI-powered repo intelligence. Auto documentation, smart PR analysis, issue triage, dependency graph analysis, code review agents — deployed and live on GitHub.com.

GitHub APIMCPAI Agents
Live on GitHub
Public
🔗
↗

21-Server MCP Ecosystem

Complete MCP server suite: file system ops, browser automation, web search, code execution sandboxes, memory bridges, API connectors, image analysis, and specialized domain tools. The backbone of M4ST.

MCP ProtocolTypeScriptPython
21 Servers · Live
Core Infra
💡

LLM Router & Cost Engine

Routes intelligently across 11 free LLM providers. Quality-aware, cost-zero, fallback-resilient. Switches providers based on task type, context length, and quality requirements.

LiteLLMRouting11 APIs
$0/month · Live
Zero Cost

TECHNICAL SKILLS

The stack

Agent Frameworks

LangGraph96%
CrewAI94%
LangChain92%
AutoGen80%
Pydantic AI85%

Languages

Python97%
TypeScript82%
JavaScript88%

Infrastructure & Tools

ChromaDBMCP ProtocolOllama FastAPIDockerLiteLLM Git / GitHubLinux CLIREST APIs VercelPostgreSQLSQLite EmbeddingsRAGVector DBs SeleniumPlaywrightBeautifulSoup

Domain Knowledge

AI OrchestrationCybersecurity OSINTSemantic Search Prompt EngineeringFine-tuning System DesignAPI Integration

BY THE NUMBERS

Impact that speaks

0

MCP Servers Built

Across all domains

0

Free LLM Providers Integrated

Routed intelligently

$0

Monthly Infrastructure Cost

Production, not demo

HOW I WORK

The process

01

Discovery & Scoping

Deep analysis of your requirements, existing systems, and workflow goals. I map every agent, every MCP server, every data flow. Nothing is assumed, everything is verified.

02

Architecture Design

Precision system architecture: agent hierarchy, orchestration logic, memory layers, LLM routing strategy, tool selection, and integration points. All blueprinted before code is written.

03

Build & Deploy

Clean, documented, production-ready code. Each agent tested in isolation, then end-to-end. Full deployment with monitoring, docs, and knowledge transfer. You own everything.

FAQ

Common questions

How do you build production AI at zero cost?
+
By routing intelligently across 11 free LLM providers — Gemini, Groq, Mistral, Cohere, HuggingFace, Together AI, and others. LiteLLM handles the routing automatically based on task type and quality requirements. The chromaDB vector store runs locally, agents run locally via Ollama for fully offline capability. No subscription to any paid AI service.
What is MCP and why does it matter?
+
MCP (Model Context Protocol) is Anthropic's open standard for connecting AI agents to external tools and data sources. Think of MCP servers as modular tool packages — file system access, web search, code execution, database connections, API bridges. I've built 21 custom MCP servers that give agents rich, reliable access to the real world without brittle prompt hacking.
Are you available for freelance work?
+
Yes — I take on a limited number of AI freelance engagements. I work on agent system architecture, MCP server development, LLM pipeline automation, semantic memory systems, and AI infrastructure consulting. I work with clients worldwide remotely. Reach out via email at m4stanuj@gmail.com.
What makes your approach different from typical AI developers?
+
Most AI developers wrap existing APIs and call it "AI". I build genuine autonomous systems — agents with real memory, real tool access, real decision-making capability. My stack is local-first which means your data never leaves your control, your system works offline, and you pay zero recurring costs. I care about the architecture, not just the demo.
Can you build for my specific industry or domain?
+
Absolutely. The MCP architecture is domain-agnostic. I've worked on OSINT intelligence systems, developer tooling, document automation, research pipelines, and more. Custom MCP servers can be built for any data source or API. Contact me to discuss your specific requirements.

LET'S BUILD TOGETHER

Want to work together?

I'm open for AI freelance work worldwide.
Based in Bareilly, India — working remotely with global clients.

📧
Email
m4stanuj@gmail.com
🐙
GitHub
m4stanuj
💼
LinkedIn
mast-anuj
✉ Send a message →
Currently accepting new AI freelance engagements · Worldwide