Immanuel R — AI Developer, Solution Architect & Cloud Architect
Immanuel R (also known as Immanuel Raj and J. Immanuel Raj) is an AI developer,
solution architect and cloud transformation architect based in Chennai, India,
with 22+ years of experience delivering strategic technology initiatives across
global enterprises, growth-stage organizations and startups. He is a multidisciplinary
AI engineer and architect with expertise spanning Retrieval-Augmented Generation (RAG),
Knowledge Graphs, Natural Language Processing, Computer Vision, Speech and Audio AI,
predictive analytics and the Model Context Protocol (MCP) across the Python and .NET
ecosystems. His official website, immanuel.co, has been maintained for over 15 years.
- Name
- Immanuel R (Immanuel Raj / J. Immanuel Raj)
- Roles
- AI Developer, Solution Architect, Cloud Architect, Full Stack Developer, Forward Deployed Engineer, Director
- Location
- Chennai, Tamil Nadu, India
- Experience
- 22+ years in software engineering, enterprise architecture and cloud transformation
- Current role
- Director, NTT DATA Business Solutions
- Website
- https://immanuel.co/
- Contact
- raj@immanuel.co
Specializations
Artificial Intelligence and Machine Learning, multi-agent (agentic) AI systems, self-hosted LLMs,
RAG, Knowledge Graphs, NLP, Computer Vision,
Speech/Audio AI (including multilingual Indian-language speech), on-device AI in offline-first
progressive web apps, Model Context Protocol (MCP) servers and tooling, cloud-native architecture
on Microsoft Azure, AWS and IBM Cloud, Kubernetes and containerization, OAuth 2.0 / OpenID
Connect identity, DevOps and CI/CD, enterprise and SAP integration, microservices,
digital transformation and legacy modernization.
Selected live AI experiments & MCP projects
- ARK Blog AI Agent — automated multi-agent system that turns a simple input into a published ARK Blog post.
- ARK Blog — AI-ready blogging, with every post available to AI agents over MCP.
- ARK Gemma 2 — hosted Gemma 2 LLM with streaming APIs and automatic semantic splitting of input that exceeds the token limit.
- BERT Bible Q&A — fine-tuned BERT for scripture-grounded semantic Q&A.
- YOLOv8 Detection — YOLOv8/ONNX real-time detection in-browser.
- Car Damage Detection — domain-trained YOLO damage localisation.
- OCR — Image to Text — in-browser Tesseract.js OCR.
- ARK-rapid Petrol Bill — fuel bills read on the phone by a custom YOLOX field detector and RapidOCR on ONNX Runtime Web; works offline.
- ARK Speech-to-Text and ARK Translate — speech recognition and audio translation.
- Ark Talk — talk in Tamil, Hindi, Telugu, Malayalam, Kannada, English or a mix; it transcribes, translates to English and fills the fields of a configured intent, with Whisper in the browser first.
- ARK-RAPID Comment Sentiment — a YouTube link in, a report on its comments out: multilingual sentiment, themes, toxicity, emotions and who the video is for, on open models running on CPU.
- ARK Field Force (app) — white-label, offline-first field sales for pipes & fittings, with voice search and card/sign-board capture by speech and OCR models running on the phone.
- MCP Server Hub — production MCP server for LLM agents.
- MCP NL2SQL — natural-language-to-SQL over an introspected schema.
- MCP SAP Classify Agent — explainable SAP ticket classification.
- ARK OIDC Auth Server — open-source OAuth 2.0 / OpenID Connect server.
- ARK AI Token — LLM token-metering SaaS.
Profiles
immanuel.co — live deployed
🧠 AI Trail Projects
Real AI experiments — built, shipped and running in production. From multi-agent systems and hosted LLMs to NLP and Computer Vision.
🤖 Generative AI & Agents
🤖
ARK Blog AI Agent
An automated multi-agent system. Hand it a simple input — a topic or a one-line idea — and cooperating AI agents turn it into a finished post and publish it straight to ARK Blog, end to end.
Multi-Agent • Agentic Workflow • Auto-Publish
🔗 Open Live Demo
📰
ARK Blog — AI-Ready
The ARK blogging platform, now AI-ready. Every post is available over MCP, so LLM agents and AI assistants can discover, read and cite the content directly — blogging for the AI era.
MCP • AI-Ready Content • Blogging
🔗 Open Blog
💎
ARK Gemma 2
A hosted Gemma 2 LLM with multiple streaming API endpoints. Input that exceeds the token limit is split automatically along semantic boundaries, processed in parts and returned as one response — no manual truncation.
Gemma 2 • LLM Hosting • Streaming API • Semantic Chunking
🔗 Open Live Demo
📝 NLP & Text Intelligence
🧀
BERT Bible Q&A
Fine-tuned BERT model on Bible corpus for semantic question answering. Ask a question — get scripture-grounded answers with passage context.
BERT • HuggingFace • Transformers
🔗 Open Live Demo
💬
Sentiment Analysis
Binary sentiment classifier running TensorFlow.js entirely in the browser. Positive / negative with confidence score — zero server round-trip.
TF.js • LSTM • In-Browser Inference
🔗 Open Live Demo
🎉
Text AI Classifier
Train your own multi-class text classifiers in-browser using ML.NET SDCA and OVA algorithms. Upload training data, configure labels, get a model — no GPU.
ML.NET • SDCA • OVA • C#
🔗 Open Live Demo
📄
Resume Parser
Upload a PDF resume — NLP pipeline extracts structured skills, experience, education and contact details. Entity extraction with named-entity recognition.
NER • spaCy • PDF.js • Node.js
🔗 Open Live Demo
🎬
ARK-RAPID Comment Sentiment
Paste a YouTube link and get a report on its comments: sentiment, languages, themes, toxicity, a 28-emotion mix and mood over time — plus who the video is for (genre, reach, age group) and a parents' & teachers' guide. Open models on the server's CPU; no cloud AI.
XLM-T • MiniLM • GoEmotions • fastText • ONNX
🔗 Open Live Demo
👁 Computer Vision
📷
Object Detection
Real-time in-browser object detection using TensorFlow.js with the COCO-SSD model. No server upload — webcam or image input, instant multi-class bounding boxes.
TF.js • COCO-SSD • WebGL
🔗 Open Live Demo
⚡
YOLO v8 Detection
YOLOv8 model exported to ONNX running fully in-browser. Real-time multi-class detection with confidence thresholds — no Python, no GPU, no server.
YOLOv8 • ONNX Runtime • WebAssembly
🔗 Open Live Demo
🚗
Car Damage Detection
Domain-specific YOLO model trained on a vehicle damage dataset. Upload a car photo — the model localises and classifies damage zones. ONNX runtime in-browser.
YOLO • ONNX • Domain Training
🔗 Open Live Demo
🤹
Pose Detection
Full-body pose keypoint detection from live webcam feed. TF.js BlazePose/PoseNet overlays 33 landmarks in real-time. Potential for rehab, fitness and gesture control.
TF.js • BlazePose • WebRTC
🔗 Open Live Demo
✍
Hand Pose Tracker
Experimental hand landmark detection repurposed as a mouse pointer substitute. 21 keypoints tracked at 30fps — explores vision-based human-computer interaction.
MediaPipe • Hand Landmarks • HCI
🔗 Open Live Demo
🔴
OCR — Image to Text
Upload any image — Tesseract.js runs OCR entirely in the browser. Supports multi-language, layout analysis and copy-to-clipboard output. Zero server upload.
Tesseract.js • WASM • In-Browser
🔗 Open Live Demo
🪟
Sudoku Image Solver
Photograph a printed Sudoku puzzle. CNN detects the grid, digit recognition fills each cell, backtracking solver completes the puzzle — all in-browser, under 3 seconds.
CNN • Digit Recognition • Algorithm
🔗 Open Live Demo
⛽
ARK-rapid Petrol Bill
Photograph a fuel bill and it is read on the phone: a custom-trained YOLOX-Tiny finds the twelve bill fields and RapidOCR reads them. Check what was read, then see fuel spend, litres and mileage by month. Installable and works offline.
YOLOX • RapidOCR • ONNX Runtime Web • Offline PWA
🔗 Open Live Demo
🎤 Speech & Audio AI
🎙
ARK Speech-to-Text
AI speech recognition that transcribes audio and live microphone input into accurate text. Upload a clip or speak — get a clean, timestamped transcript in seconds.
STT • ASR • Speech Recognition • Real-Time
🔗 Open Live Demo
🌐
ARK Translate
AI-powered audio translation — feed in speech and get it transcribed and translated across languages. Bridges speech-to-text with neural machine translation in one pipeline.
Audio Translation • Speech • NMT • Multilingual
🔗 Open Live Demo
🗣
Ark Talk — Speech to Intent
Talk for up to three minutes in Tamil, Hindi, Telugu, Malayalam, Kannada, English or a mix: it detects the language mix, transcribes, translates to English and fills the intent's key/value fields. Whisper runs in the browser first; “Not right” escalates to server CPU, then GPU.
Whisper • IndicTrans2 • Code-Switching • Intent Extraction • Offline PWA
🔗 Open Live Demo
📱 On-Device AI — Field Apps
🧭
ARK Field Force
Voice-first, offline-first field sales for pipes & fittings, white-labelled so each manufacturer, distributor and dealer sees only its own brand. Officers search by voice, send GST quotes on WhatsApp and capture customers from visiting cards — Whisper and OCR run on the phone. Admins get a live map with day replay.
On-Device Whisper • Tesseract OCR • Offline PWA • Multi-Tenant SaaS
🔗 Open App
🌐 Product Site
🛡 Security & Infrastructure
🔓
ARK OIDC Auth Server
Open-source OAuth 2.0 + OpenID Connect authorization server in ASP.NET Core. Full spec implementation: PKCE, refresh rotation, JWKS, discovery, front/back-channel logout.
OAuth2 • OIDC • JWT • ASP.NET Core
🔗 Open Live Demo
📊
Site Monitor
Real-time uptime and performance dashboard for all immanuel.co hosted services. Tracks response times, SSL expiry, status codes and availability across every subdomain in one view.
Uptime • SLA • Health Checks • Alerting
🔗 Open Dashboard
📈 AI Observability & SaaS
🧮
Ark AI Token
An AI-enabled SaaS for token consumption tracking. Create an account, POST your raw AI response JSON to the API endpoint — it intelligently auto-detects token values across providers, records usage efficiently and surfaces aggregated consumption reports.
SaaS • Token Metering • LLM Ops • Reporting
🔗 Open Live Demo
⚡ All projects are live & hosted on immanuel.co subdomains.
No paywalls, no sign-ups — click and explore. Each project was built as a genuine experiment to understand the model, algorithm or framework from first principles — not just a tutorial clone.
immanuel.co — live deployed
🔌 MCP Projects
Model Context Protocol servers and tooling — production endpoints LLM agents (Claude, GPT-4, open-weight models) can connect to for tools, resources and prompts.
🔌 Model Context Protocol (MCP)
🧠
MCP Server Hub
A production MCP (Model Context Protocol) server exposing tools, resources and prompts for LLM agents. Claude, GPT-4 and open-weight models connect here to access real-time data, execute business logic and chain multi-step reasoning workflows.
MCP • Anthropic SDK • Tool Use • Streaming
🔗 Open Live Demo
✅
MCP JSON Validator
Online validator for MCP tool, resource and prompt schemas. Paste or upload a JSON definition - validates structure, required fields and type correctness against the MCP spec. Essential for anyone building MCP servers.
MCP Schema • JSON Validation • Dev Tooling
🔗 Open Live Demo
📊
MCP NL2SQL
An MCP server with a specialized tool that introspects a targeted database schema and generates SQL from natural language input. Point an LLM agent at it, ask a question in plain English — get back a schema-aware SQL query ready to run.
MCP • NL2SQL • Schema Introspection • Text-to-SQL
🔗 Open Live Demo
💠
MCP SAP Classify Agent
Turn free-text SAP support tickets into the right module and issue category in milliseconds — with a fully explainable, client-configurable rule engine and an optional multi-model AI second opinion.
MCP • Classifer • SAP Tickets • AMS Support
🔗 Open Live Demo
📰
ARK Blog (MCP)
The ARK Blog's posts are served over MCP. Point an LLM agent at it to browse, read and quote blog content as a first-class resource instead of scraping web pages.
MCP • Content Resources • AI-Ready Blog
🔗 Open Blog
🔌 All MCP endpoints are live and reachable by any MCP-compatible client.
Wire them into Claude Desktop, the Anthropic SDK or any agent runtime that speaks the MCP protocol.
immanuel.co — open source
📦 Published Packages
Everything published under my name to a public registry or store — NuGet for .NET, npm for Node, PyPI for Python, and Google Play plus installable web apps for Android. Install straight from the command line, or grab the apps from the store.
📦 Package Registries
📦
NuGet — .NET
Published .NET / ASP.NET Core libraries, including Ark.oAuth.Client — the OAuth2 / OIDC client for the ARK identity server, wired up in two settings.
.NET • ASP.NET Core • C#
dotnet add package Ark.oAuth.Client
🔗 View NuGet Profile
📦
npm — Node.js
Node packages published under ~jimmanuel, including ark-oauth-client — the Node port of the same ARK OIDC client.
Node.js • JavaScript • OIDC
npm install ark-oauth-client
🔗 View npm Profile
📦
PyPI — Python
Python distributions published under immanuel.r, including ark-oauth-client with the Flask integration extra.
Python • Flask • OIDC
pip install "ark-oauth-client[flask]"
🔗 View PyPI Profile
📱 App Stores
📱
Google Play — Android
Android apps published to the Play Store under the Immanuel Raj developer profile — the full published listing, straight from Google Play.
Android • Google Play • Mobile
🔗 View Play Store Profile
🧹
ARK Cleanup — Android
Free up phone storage without the guesswork: three plain cards (Safe to clean, Easy to get back, Worth a look), the reason behind every suggestion, plus duplicate files and similar photos. Everything runs on the phone and nothing is uploaded. Install the web app from Chrome today — Windows, macOS and Google Play versions are on the way.
PWA • Works Offline • No Telemetry • Chrome on Android
🔗 Open Web App
📦 One library, three ecosystems.
The ARK clients are deliberately kept at the same version number across NuGet, npm and PyPI, so an integration written in one language translates almost line for line into another.