AI & IoT Engineer at Suraj Informatics · Mumbai

From field signals to intelligent systems.

I'm Anand Kumar Maurya. I build the whole chain: sensors and edge boards, the backends that move their data, and the AI agents that make sense of it. Industrial sites by day, open-source tools by night.

Explore my work GitHub
years shipping to production
4+
years shipping to production
industrial systems delivered
15+
industrial systems delivered
open-source projects
30+
open-source projects
node homelab for edge AI
6
node homelab for edge AI

ops agent

128 devices · 6 nodes · MCP

live

industrial iot
Java · Spring BootGoKotlinRustMQTT · EMQXRFID · ANPRGPS / DGPSUART · CAN · I2CRockchip NPUllama.cppMCP serversSpring AI · RAGPostgreSQLRabbitMQ · RedisKeycloak · OIDCDocker · ProxmoxOpenWrtReact · Next.js

Selected work

Things I've built because I wanted them to exist.

Open-source tools for edge AI, AI agents and self-hosting. Most started as a problem in my own homelab, and I use every one of them.

Self-hosted platforms
Multi-cluster · REST + MCP

Ferrum

Fleet control for Proxmox VE.

One dashboard for every cluster and standalone node: live inventory, backups, HA, firewall, alerting, a REST API and a built-in MCP server so AI clients can operate the fleet.

GoReactTypeScriptProxmox APIMCP 13
Edge AI & hardware
5 Rockchip SoCs

ggml-rknpu2

Run llama.cpp on the Rockchip NPU.

A ggml backend that offloads LLM inference to the NPU on RK3588, RK3576, RK3566, RK3568 and RK3562 boards, with a one-line installer and a simple llama-npu command.

C++ggmlllama.cppRKNPU2GGUF
Edge AI & hardware
100% on-device

Vervan Chat

A private AI assistant that lives on your phone.

Offline-first Android assistant: chat, voice, document Q&A and screen understanding with models that run on-device. Remote OpenAI-compatible models are optional.

KotlinJetpack Composellama.cppLiteRT 2
AI agents & MCP
Keys never reach the model

ssh-mcp

Audited SSH access for AI assistants.

Gives AI assistants controlled SSH, SFTP and tunnel access to servers without ever exposing passwords or keys to the model. Portable Go and Java builds.

GoJavaMCPSSH
Edge AI & hardware
ANPR on a $10 board

Luckfox PicoCam

Open IP-camera firmware with on-board AI.

Firmware for the Luckfox Pico family (RV1103/RV1106): web portal, RTSP/ONVIF streaming, recording and NPU-powered detection such as ANPR, on a board the size of a stamp.

C++RV1106RTSPONVIF
Edge AI & hardware
GPU · NPU · Apple · CPU

SayWho

Who spoke when, in any recording.

Self-hosted speaker diarization web app and REST API. Runs on an NVIDIA GPU, a Rockchip NPU, Apple Silicon or plain CPU.

PythonFastAPICUDARKNN
Apps & tools
GUI · TUI · CLI

StewardGet

A fast, modern download manager.

Multi-connection downloads, safe resume, browser capture and link refresh. Desktop app, TUI and CLI for Windows, macOS and Linux, with no telemetry.

RusteguiratatuiBrowser extension 1
Self-hosted platforms
Desktop · server · browser

Whittle

Parametric 3D CAD, open source.

Sketch, extrude, fillet and export STEP. Built on OpenCascade; runs as a desktop app, a self-hosted server or in the browser.

TypeScriptOpenCascadeElectronWebGL 1
Browse all 30 projects

In production · Suraj Informatics

Systems that run gates, yards and warehouses, every single day.

Client work, so the code stays private. Here is the problem, the approach and what changed. I led delivery on most of these while writing a good share of the code myself.

The problem

Operations teams often need to inspect multiple dashboards, logs, alerts, and records to understand why a device failed or what changed before an issue.

What I built

A monitoring platform that combines MQTT telemetry, Spring Boot services, event history, RAG, local LLMs, and AI-assisted workflows to help users query device status, logs, alerts, and operational records in natural language.

Impact

40–60%

  • less manual alert investigation
  • natural-language search across 30+ device events, alerts, and operational records.
JavaSpring BootSpring SecurityKeycloakRabbitMQRedisPostgreSQLMQTT

How it's put together · simplified, client details removed

Devices

  • Sensors and gateways

  • EMQX MQTT broker

Pipeline

  • Spring Boot services

  • RabbitMQ

    durable queues

  • Redis

    live state

  • PostgreSQL

    events, alerts, records

AI layer

  • RAG index

    logs, alerts and history

  • Local LLM

    runs on-prem

  • Spring AI

    prompts and tools

People

  • Ask in plain language

  • Dashboards and alerts

  • Keycloak sign-in

Answers come from the platform's own telemetry and records, not from the model's memory.

How I work

How a signal becomes a decision.

Every system I build follows the same path. I'm comfortable at every step, which matters, because most bugs hide in between.

01 / Sense

Sensors, readers and cameras

Field hardware is noisy: readers drop off, scales drift, GPS jumps. I work right at the wire, validating and de-duplicating signals before they can corrupt anything downstream.

Tools

RFIDANPRGPS / DGPSUWBUART · RS-485CAN · I2C · SPI

Where I've done it

  • Gate Operating System
  • Weighbridge automation
  • Luckfox PicoCam

The lab

Where I break things on purpose.

A six-node homelab on one routed network. It is my test bench for edge AI, infrastructure and the self-hosted tools above, and it is operated partly by my own MCP agents.

OpenWrt router

WAN · VPN · firewall

Switched fabric

10 endpoints

Proxmox cluster

4× SBC nodes + 2 mini PCs

  • Raspberry Pi 5 ×3
  • Radxa A7A
  • Mini PC ×2

Local AI

LLMs, OCR and speech, offline

  • Mac mini M4
  • Radxa ROCK 4D · RK3576
  • RK3588 NPU boards

Storage

Shared storage and backups

  • CM3588 NAS
  • TrueNAS

6

Proxmox hosts

Docker and LXC workloads, with VMs managed by Ferrum and Proxmox MCP.

Running on it

  • OpenWrt routing & firewall
  • SSO, VPN & reverse proxy
  • Beszel monitoring + MCP agents

About

Happiest where hardware meets software.

I'm an engineer at Suraj Informatics in Mumbai, where I lead a team of six building industrial IoT and AI systems: gate automation, yard and fleet tracking, weighbridges, warehouses and document pipelines.

The part I enjoy most is the space between layers, where a serial frame from a reader becomes an event, the event becomes a record, and the record becomes an answer someone can act on.

Outside work I build open-source tools, run a six-node homelab and spend far too long getting more tokens per second out of tiny NPUs.

Design for failure

Assume devices can disconnect, networks can fail, queues can back up, and users can make mistakes. Build systems that recover, explain, and keep operating.

Make systems observable

Logs, audit trails, dashboards, status history, and alerts are not optional. They are what make production systems maintainable.

Keep architecture practical

I prefer architecture that helps teams ship, debug, scale, and maintain systems instead of adding complexity for its own sake.

Use AI with control

AI should assist people, not blindly replace decisions. I prefer human-in-the-loop workflows, controlled tool access, and explainable outputs.

  1. Now

    Edge AI, Agentic AI, and Practical Automation

    Building systems that combine backend platforms, IoT telemetry, infrastructure, local AI, and agentic workflows into useful engineering solutions.

  2. 2024-2025

    AI-Enabled Systems and Infrastructure

    Expanded focus into Spring AI, MCP servers, RAG, local LLMs, and AI-assisted monitoring while continuing backend and IoT platform work.

  3. 2023-2024

    Lead + Developer Responsibilities

    Took ownership of project delivery, backend architecture, team coordination, debugging, deployment support, and production issue resolution.

  4. 2022-2023

    Industrial IoT and Backend Systems

    Worked on automation systems involving device communication, backend workflows, dashboards, reports, and production deployments.

  5. Feb 2022

    Joined Suraj Informatics

    Started working on backend and IoT systems, building Spring Boot services and integrating real-world devices with software platforms.

  6. Nov 2021

    Started Freelance Development

    Built a complete business website with backend functionality and payment gateway integration, gaining early experience in full-cycle delivery.

  7. 2017 – 2022

    MCA & B.Sc. IT

    Master of Computer Applications (CGPA 9.20) and B.Sc. Information Technology (CGPA 8.82), Mumbai. Microsoft Azure Fundamentals (AZ-900).

Backend

  • Java
  • Spring Boot
  • Go
  • Microservices
  • PostgreSQL
  • RabbitMQ
  • Redis
  • Keycloak · OAuth2

IoT & hardware

  • MQTT · EMQX
  • RFID
  • ANPR
  • GPS / DGPS
  • UWB
  • UART · CAN · I2C
  • ESP32
  • Raspberry Pi

AI & edge

  • Spring AI
  • MCP servers
  • RAG
  • Tool calling
  • llama.cpp
  • Rockchip NPU
  • Local LLMs
  • OCR · Speech

Infra & apps

  • Linux
  • Docker
  • Proxmox
  • OpenWrt
  • Nginx
  • Kotlin · Compose
  • React · Next.js
  • Rust

Contact

Have an interesting problem?

I like hard problems where software meets the physical world: devices, edge hardware, data pipelines and AI. If you are building something in that space, or just want to compare notes, my inbox is open.

anand34577@gmail.com
GitHub@anand34577LinkedInanand-k-mauryaMedium@anand34577

Send a message

Ideas, questions about a project, or something you're building. I usually reply within a couple of days.

Easier to talk it through?

Book a call