DEV Community: Akash Kumar

AuthAgent: Secure AI Assistant with Auth0

Akash Kumar — Thu, 09 Oct 2025 00:47:31 +0000

What I Built
AuthAgent is a secure AI assistant designed to help users manage personal tasks, retrieve information, and interact with third-party services—all while ensuring robust identity protection. The agent solves the problem of secure delegation: how can users trust an AI to act on their behalf without compromising their identity or data?

Key features:

Natural language interface for task automation

Secure authentication and authorization using Auth0

Role-based access control for sensitive actions

Session management and token refresh for long-lived interactions

Demo
GitHub Repository

Demo Video

How I Used Auth0 for AI Agents
Auth0 was the backbone of my agent’s identity layer. Here's how I integrated it:

Agent Authentication: Each AI agent instance authenticates using Auth0's Machine-to-Machine flow, ensuring secure API access.

User Delegation: Users log in via Auth0’s Universal Login, and their access tokens are securely passed to the agent for delegated actions.

Role-Based Access Control: Auth0 roles determine what the agent can do on behalf of each user (e.g., read calendar vs. send emails).

Token Management: The agent handles token refresh and expiration gracefully, ensuring uninterrupted sessions.

Lessons Learned and Takeaways
Building AuthAgent taught me a ton about secure agent design:

Challenge: Managing token lifecycles and securely storing credentials in agent memory was tricky. I had to implement encrypted local storage and strict access policies.

Insight: Auth0’s extensibility made it easy to plug into my agent’s lifecycle, especially with Rules and Actions for custom claims.

Advice: Don’t treat AI agents like traditional apps—think of them as semi-autonomous users. Secure them accordingly!

EduTrack: A Smarter Dashboard for Students and Educators

Akash Kumar — Fri, 26 Sep 2025 04:00:44 +0000

What I Built
I built EduTrack, a responsive dashboard for students and educators to track assignments, grades, attendance, and progress in real time. The app solves the problem of fragmented academic data by providing a unified interface that’s both visually intuitive and highly interactive.

EduTrack empowers students to stay organized and helps educators monitor class performance with ease.

Demo
Live App on Vercel

KendoReact Components Used
Chart – for visualizing grade trends and attendance over time

Scheduler – to manage class schedules and assignment deadlines

Grid – for displaying student data, sortable and filterable

Form – for submitting assignments and feedback

AutoComplete – for quick student lookup

TabStrip – to organize dashboard sections cleanly

[Optional: Code Smarter, Not Harder prize category] AI Coding Assistant Usage
I used Copilot to:

Generate dynamic chart configurations based on user input

Refactor complex grid logic for better performance

Draft reusable form components with validation

Suggest accessibility improvements for keyboard navigation

[Optional: RAGs to Riches prize category] Nuclia Integration
EduTrack integrates Nuclia’s RAG capabilities to:

Index uploaded lecture notes and PDFs

Enable semantic search for key concepts

Provide AI-generated summaries and flashcards from class materials

This turns static content into interactive learning tools, enhancing retention and engagement.

ClassMate AI: Your Smartest Study Buddy for the School Year

Akash Kumar — Fri, 26 Sep 2025 03:56:54 +0000

What I Built
I created ClassMate AI, a multi-agent assistant designed to help students manage their academic life more efficiently. It solves common back-to-school challenges like organizing assignments, summarizing lecture notes, generating study quizzes, and even suggesting personalized learning resources based on performance trends.

ClassMate AI acts as a virtual study buddy that understands your schedule, adapts to your learning style, and keeps you on track throughout the semester.

Category
Student Success

Educator Empowerment

Crazy Creative

Demo
Live App on Heroku

Source Code on GitHub

Screenshots & Demo Video:

How I Used Heroku AI
ClassMate AI uses:

Heroku Managed Inference to run a custom GPT-based summarization agent for lecture notes.

pgvector for Heroku Postgres to store and retrieve semantic embeddings of notes and resources.

Model Context Protocol (MCP) to coordinate agents: one for note summarization, one for quiz generation, and one for resource recommendation.

Agents communicate via shared context and trigger each other based on user actions (e.g., uploading notes triggers summarization, which then triggers quiz generation).

Technical Implementation
Multi-agent architecture using MCP:

SummarizerAgent: Extracts key points from uploaded notes.

QuizAgent: Generates multiple-choice questions based on summaries.

TutorAgent: Suggests resources using vector similarity from pgvector.

Tech Stack:

Frontend: React + Tailwind

Backend: Node.js + Express

AI: OpenAI GPT + Heroku Managed Inference

Database: Heroku Postgres with pgvector

Challenges Solved:

Coordinating agent workflows using MCP

Embedding and querying semantic data efficiently

Ensuring fast inference with Heroku's AI infrastructure

ZK-Chat — A Private Messaging DApp

Akash Kumar — Mon, 25 Aug 2025 07:10:08 +0000

*This is a submission for the [Midnight Network "Privacy First" Challenge]

ZK-Chat — A Private Messaging DApp
What I Built
ZK-Chat is a decentralized messaging app that enables encrypted, anonymous communication using zero-knowledge proofs for identity and message integrity. It solves the problem of surveillance and metadata leakage in traditional messaging platforms.

Demo

Encrypted chat interface

ZK proof of sender authenticity

Message integrity validation

How I Used Midnight's Technology
Compact Language: Built circuits for sender identity and message integrity

MidnightJS: Used to generate proofs and encrypt messages

Smart Contracts: Store encrypted messages and validate sender proofs

Data Protection as a Core Feature
Messages are end-to-end encrypted

Sender identity is verified via ZK proofs, not exposed

No metadata (IP, location, etc.) is stored

UI includes “Privacy Bubble” showing encryption status

Set Up Instructions / Tutorial
Clone repo and install dependencies

Compile ZK circuits for messaging

Deploy contracts to Midnight testnet

Start frontend and simulate chat with mock users

View encrypted messages and proof validation

Smart Data Sync Agent

Akash Kumar — Fri, 15 Aug 2025 15:44:18 +0000

This is a submission for the AI Agents Challenge powered by n8n and Bright Data

What I Built
I created a Smart Data Sync Agent that automates real-time data integration between on-premises systems and cloud services using Azure Integration Services, n8n, and Bright Data. This agent solves the problem of fragmented enterprise data by securely bridging legacy systems with modern cloud workflows—without manual intervention.

Key capabilities:

Monitors on-prem SQL Server for new entries via Azure Data Gateway

Uses n8n to orchestrate workflows and trigger actions

Enriches data using Bright Data’s Verified Node (e.g., company info, market data)

Publishes results to cloud apps like SharePoint, Teams, or Power BI

Demo
Demo Video on YouTube

neon Workflow
View Workflow JSON on GitHub Gist

Technical Implementation
System Instructions
The agent is configured to:

Poll on-prem data using Azure Logic Apps and Data Gateway

Trigger n8n workflows for transformation and enrichment

Use Bright Data to fetch verified external data

Output results to cloud dashboards or messaging platforms

Model Choice
n8n handles orchestration and logic

Azure handles secure data access and messaging

Bright Data provides real-time verified data enrichment

Memory
Stateless design with contextual memory via n8n’s built-in data store

Temporary caching for deduplication and retry logic

Tools Used
Azure Integration Services: Logic Apps, Service Bus, Data Gateway

n8n: Workflow automation and orchestration

Bright Data Verified Node: External data enrichment

Power BI / SharePoint / Teams: Output destinations

Bright Data Verified Node
I used Bright Data’s Verified Node to:

Enrich company records with verified business info

Validate contact details before syncing to CRM

Pull market trends for financial dashboards

This added a layer of intelligence and trust to the data pipeline, making the agent useful for sales, finance, and operations teams.

Journey
This challenge pushed me to explore hybrid integration scenarios and real-time data enrichment. Key milestones:

Setting up Azure Data Gateway for secure on-prem access

Designing modular n8n workflows with error handling

Integrating Bright Data for verified external insights

Testing across multiple cloud endpoints

Challenges overcome:

Handling intermittent connectivity with retry logic

Mapping legacy data formats to modern APIs

Ensuring security and compliance in hybrid workflows

What I learned:

The power of combining low-code tools like n8n with enterprise-grade services

How Bright Data can elevate the quality of AI-driven decisions

That hybrid integration is not just possible—it’s scalable and elegant