I’m Saurabh Jain, a Full-Stack AI Engineer and founder of SMAKG.com. I specialize in custom backend architecture, 100% serverless RAG systems, and manual orchestrating layers - no black-box wrapper tech. Previously, I was a Founding Engineer at Austria’s largest legal AI startup. I’m a top 3% open-source contributor with code merged in Cal.com and Twenty CRM, and the creator of KRAG, a serverless retrieval system architected to outperform NotebookLM in privacy and speed. Checkout my portfolio: _Website not available. Sign in: https://www.twine.net/signup_

Saurabh Jain

PRO

I’m Saurabh Jain, a Full-Stack AI Engineer and founder of SMAKG.com. I specialize in custom backend architecture, 100% serverless RAG systems, and manual orchestrating layers - no black-box wrapper tech. Previously, I was a Founding Engineer at Austria’s largest legal AI startup. I’m a top 3% open-source contributor with code merged in Cal.com and Twenty CRM, and the creator of KRAG, a serverless retrieval system architected to outperform NotebookLM in privacy and speed. Checkout my portfolio: _Website not available. Sign in: https://www.twine.net/signup_

Available to hire

I’m Saurabh Jain, a Full-Stack AI Engineer and founder of SMAKG.com. I specialize in custom backend architecture, 100% serverless RAG systems, and manual orchestrating layers - no black-box wrapper tech. Previously, I was a Founding Engineer at Austria’s largest legal AI startup. I’m a top 3% open-source contributor with code merged in Cal.com and Twenty CRM, and the creator of KRAG, a serverless retrieval system architected to outperform NotebookLM in privacy and speed.

Checkout my portfolio: Website not available. Sign in: https://www.twine.net/signup

See more

Language

English
Fluent

Work Experience

Founder at SMAKG.com
February 1, 2026 - Present
Leading SMAKG.com, defining product vision, technical strategy, and go-to-market. Developing flagship products and client solutions.
Founding Engineer at Buildway.ai
October 1, 2025 - March 1, 2026
Engineered a multi-agent system and improved RAG citation accuracy by 90%. Built billing infrastructure and an AI chat system. Executed a zero-downtime migration.

Education

Bachelor of Science in Data Science at Indian Institute of Technology, Madras
January 1, 2025 - January 1, 2029

Qualifications

Docker Certification
January 1, 2025 - March 27, 2026
GitHub Actions Workshop: CI/CD Pipelines
January 1, 2025 - March 27, 2026

Industry Experience

Software & Internet, Government
    uniE608 StealthNode | AI SOC Analyst
    StealthNode: Automated, AI-Driven SOC Analyst for Local Defense Live Architecture Breakdown: Watch on YouTube 📌 Project Overview When a local AI agent is compromised, privilege escalation and data exfiltration can happen in exactly 3 seconds—long before a human ever notices 00:00:00]. To solve this, I engineered StealthNode, an automated, AI-driven Security Operations Center (SOC) analyst designed to react to local threats in seconds. It drops the massive overhead of a traditional SOC directly onto a local machine, silently monitoring and neutralizing attacks on AI agents before the damage is done 00:00:47]. 🏗️ Core Architecture & Technical Stack StealthNode is built on a highly secure, custom event-driven architecture designed for low latency and absolute system safety 00:01:27]. Real-Time Telemetry: A lightweight powershell installer deploys a Fast MCP and a Wazuh agent locally, persisting through reboots 00:01:36]. The agent continuously streams logs to a dedicated Wazuh server, where custom-written rule sets filter noise and detect legitimate threats 00:01:52]. Serverless Orchestration: Upon threat validation, the event is pushed to a queue that instantly spins up an isolated, serverless Modal sandbox 00:01:59]. This ensures the defensive compute is entirely segregated from the potentially compromised host machine. Secure Tunneling & Analysis: The system injects a Fast MCP URL directly into the sandbox via a completely secure, custom Cloudflare tunnel network 00:02:07]. A Claude-powered security agent is then deployed inside the sandbox to analyze the telemetry and execute countermeasures. Strictly Controlled Execution: To ensure the AI doesn't nuke the host operating system, the MCP is restricted to three highly specific tools: executing osquery, disabling compromised users, and running a strictly whitelisted set of system commands 00:02:25]. 🔒 Future-Proofing & Roadmap StealthNode is currently live in beta for Windows, continuously proving its ability to intercept zero-day behavior 00:03:59]. The architecture is actively being expanded to include specialized Security Small Language Models (SLMs) specifically designed to detect and block prompt injection and infiltration attempts against local AI models 00:04:14]. * *Architected for complex, low-latency defense. I build highly secure, event-driven infrastructure from scratch.*
    uniE608 KRAG | I Built the World's First 100% Private RAG Platform
    KRAG: The World's First 100% Serverless & Private RAG Platform Live Architecture Breakdown: Watch on YouTube Source Code: GitHub Repository 📌 Project Overview The greatest bottleneck with enterprise AI today is the tradeoff between data privacy and decent results; companies are forced to send their intellectual property to external foundational models. To solve this, I engineered and open-sourced KRAG under the GPL3 license. It is a fully serverless, locally-hosted Retrieval-Augmented Generation (RAG) platform that guarantees zero data leakage while delivering state-of-the-art multi-modal performance. 🏗️ Core Architecture & Technical Stack Unlike standard wrapper APIs, KRAG is built on a highly optimized, custom multi-modal pipeline designed for maximum security and zero idle costs. Serverless Compute Engine: The entire heavy lifting runs on Modal serverless GPUs. The infrastructure ensures you only pay for active execution time, completely eliminating expensive idle server costs. Advanced Multi-Modal Parsing: The pipeline utilizes the Marker PDF parser to accurately extract dense document content into Markdown, heavily outperforming basic parsers. Furthermore, it integrates Florence-2 for high-fidelity image summarization, ensuring the RAG agent contextualizes visual data natively. Precision Retrieval & Reranking: Implements parent-child chunking combined with BGE-M3 embeddings. The retrieval pipeline optimizes for vector and keyword search, passing results through an mxbai reranker to guarantee only the highest quality chunks reach the citation extractor. Sliding Window Memory: Maintains an 8,000-token context window, dynamically preserving 4,000 tokens exactly as they are while summarizing the remaining 4,000 to maintain deep conversational context without blowing up compute. 🔒 Enterprise-Grade Security Security is baked into the foundation. The platform utilizes 3-tier encryption, ensuring that embeddings are inherently encrypted at rest. With encryption toggled on, all parsed data saved to the database is completely secured, keeping you as the absolute master of your own proprietary data. * *Engineered to push the limits of automated, secure retrieval. I write manual orchestrating layers—no black boxes.*