Eklavya Platform

500K IT jobs at risk. Upskilling costs ₹5L. Eklavya: Free AI learning, 9 virtual mentors, 16-model comparison. Learn by doing, not paying. Powered by Claude 4.5. 98% cheaper.

Comment

Inspiration

In the ancient Indian epic Mahabharata, Eklavya learned archery by practicing in front of a statue of the legendary master Dronacharya. Without direct access to the guru, he became one of the greatest archers through dedication and self-learning.

Today, you can't access tech legends directly. But their AI-powered virtual versions can be YOUR mentors — 24/7.

The timing couldn't be more urgent. India's IT sector is facing its biggest workforce disruption in history:

  • TCS is laying off 2% of its workforce (~12,000+ employees) — the largest layoffs in company history
  • Up to 500,000 IT jobs are at risk in the next 3 years due to AI automation
  • 68% of professionals expect their roles to be partially or fully automated within 5 years
  • 4 in 10 workers fear their current skills will soon be obsolete

The message from industry leaders is clear: “Reskill or risk redundancy.”

But here's the problem: Traditional AI upskilling is expensive (₹2-5 Lakhs) and slow (12-18 months). Workers facing job displacement can't afford either.

That's why I built Eklavya — a free, interactive AI learning platform that makes world-class AI education accessible to everyone, immediately.

What it does

Eklavya combines education, mentorship, and model comparison into one powerful platform:

1. Interactive Learning

Learn AI concepts through hands-on lessons with real-time feedback:

  • Prompt Engineering Fundamentals: Compare basic vs well-structured prompts side-by-side
  • RAG (Retrieval Augmented Generation): Paste your own documents and see how context improves AI responses
  • Claude API Integration: Real-time token tracking, cost analysis, and production-ready examples

2. Virtual Mentorship

Chat with AI personas of legendary tech leaders:

  • Jeff Dean (Google Senior Fellow) — Distributed Systems, ML at Scale
  • Andrej Karpathy (Ex-Tesla AI Director) — Deep Learning, Neural Networks
  • Arpit Bhayani (System Design Expert) — Backend, Scalability
  • Linus Torvalds (Creator of Linux & Git) — Open Source, Kernel Development
  • Kent Beck (Creator of XP & TDD) — Software Craftsmanship
  • Paul Graham (Y Combinator Co-founder) — Startups, Product
  • Reshma Saujani (Girls Who Code Founder) — Diversity in Tech
  • Steve Ballmer (Former Microsoft CEO) — Business & Tech Strategy
  • Shigeru Miyamoto (Creator of Mario & Zelda) — Game Design, UX

3. Model Comparison Arena

Compare up to 16 AI models in real-time with a single prompt:

  • Live API calls to Anthropic, OpenAI, Google, Meta, Mistral, DeepSeek, Perplexity, Grok, Sarvam, and Ollama
  • Side-by-side responses streaming in real-time
  • Performance charts showing response time, cost, and token usage
  • Data-driven decisions for choosing the right model for your use case

How we built it

Tech Stack

  • Frontend: Next.js 14, Tailwind CSS, Framer Motion, Monaco Editor
  • Backend: Claude Sonnet 4.5, Next.js API Routes
  • AI Providers: Anthropic, OpenAI, Google, Groq, Ollama, and more

Development Process

  1. Platform Architecture: Built on Next.js 14 with App Router and React Server Components for optimal performance
  2. AI Integration: Implemented streaming responses for all 16 models using their respective SDKs
  3. Mentor Personas: Designed 9 unique system prompts based on each legend's teaching philosophy and communication style
  4. Interactive Lessons: Created hands-on exercises with live code editors, RAG context input, and real-time API tracking
  5. Real-time Charts: Built performance visualization using Chart.js for latency, cost, and token comparison

Challenges we ran into

1. Multi-Provider API Integration

Each AI provider has different SDKs, authentication methods, and response formats. Creating a unified interface that could handle 16 models from 10 different providers required:

  • Custom adapter patterns for each provider
  • Error handling for rate limits, timeouts, and API failures
  • Streaming response normalization across different formats

2. Real-time Performance Comparison

Building live charts that update as model responses stream in was technically complex:

  • Token counting had to happen in real-time during streaming
  • Cost calculation required provider-specific pricing logic
  • Latency measurement needed precise timing for each API call
  • Chart animations had to be smooth while data updated continuously

3. Token Confusion

The biggest user education challenge was explaining tokens vs characters. Users would paste 100-word documents and wonder why they got charged for “2,000 tokens.” I solved this by:

  • Adding a live token visualization widget
  • Creating an interactive token calculator
  • Explaining tokenization with real-time examples

4. RAG Context Management

Making RAG (Retrieval Augmented Generation) understandable for beginners was challenging:

  • How much context is too much?
  • How to structure context for best results?
  • When does RAG help vs hurt?

I addressed this with an interactive lesson where users paste their own documents and immediately see how context affects responses.

5. Build and Deployment

The platform had 50+ ESLint warnings (unescaped apostrophes, unused imports, etc.) that blocked production builds. I had to balance:

  • Shipping quickly for the hackathon deadline
  • Maintaining code quality standards
  • Ensuring production deployment success

I disabled strict linting for builds while keeping it active in development, allowing deployment without compromising dev experience.

Accomplishments that we're proud of

1. Educational Impact

Created a platform that reduces AI learning costs by 98% compared to traditional courses:

  • Traditional courses: ₹2-5 Lakhs for 12-18 months
  • Eklavya: Free platform + ₹100-500/month API costs

2. Virtual Mentorship at Scale

Built AI personas of 9 legendary tech leaders who are typically inaccessible:

  • No scheduling conflicts
  • No hourly fees (vs ₹2-6K/hour on platforms like Topmate)
  • 24/7 availability
  • Unlimited sessions

3. Real Production Tools

This isn't a toy — it's production-ready:

  • 16 real AI models with live API calls
  • Real-time token tracking for cost optimization
  • Actual streaming responses (not simulations)
  • Monaco Editor for code examples
  • RAG implementation with custom context

4. Immediate Practical Value

Users can:

  • Test different prompts and see results instantly
  • Compare models before committing to an API provider
  • Learn prompt engineering by doing, not watching videos
  • Get mentorship from virtual versions of tech legends

What we learned

1. Claude as a Development Partner

Building with Claude Sonnet 4.5 taught me how powerful AI-assisted development can be:

  • Rapid prototyping of complex features
  • Real-time debugging and optimization
  • System prompt engineering for mentor personas
  • Context management for long conversations

2. Prompt Engineering is an Art AND Science

Creating effective prompts requires:

  • Clarity: Specific instructions work better than vague requests
  • Context: Providing relevant background improves accuracy
  • Structure: Using headings, lists, and examples enhances output
  • Iteration: Testing and refining prompts is essential

3. RAG is a Game-Changer

Retrieval Augmented Generation solves the “knowledge cutoff” problem:

  • AI can answer questions about proprietary documents
  • Context injection improves accuracy dramatically
  • But there's a token cost vs accuracy tradeoff to manage

4. Model Diversity Matters

Different models excel at different tasks:

  • Claude: Nuanced reasoning, code generation
  • GPT-4: General knowledge, creative writing
  • Gemini: Multilingual tasks, data analysis
  • Groq: Ultra-fast inference
  • DeepSeek: Cost-effective coding tasks

Having a comparison tool helps users make data-driven decisions instead of relying on marketing claims.

5. Education Must Be Hands-On

Reading about prompt engineering vs actually trying it are completely different experiences. The interactive lessons where users can paste their own content, modify prompts, and see real-time results are far more effective than passive tutorials.

What's next for Eklavya Platform

Immediate (Next 2 Weeks)

  • Fix build warnings for production-grade code quality
  • Add more mentor personas: Douglas Crockford (JavaScript), Rich Hickey (Clojure/FP), Tim Berners-Lee (Web)
  • Improve mobile responsiveness for on-the-go learning
  • Add user authentication to save chat history and learning progress

Short-term (1-2 Months)

Advanced Learning Modules:

  • Multi-turn prompt patterns
  • Agent orchestration with Claude
  • Function calling and tool use
  • Fine-tuning vs RAG tradeoffs

Other Features:

  • Code Execution Sandbox: Let users run AI-generated code safely in the browser
  • Mentor Memory: Allow mentor chats to remember previous conversations across sessions
  • Model Fine-tuning Guide: Interactive lesson on when and how to fine-tune models

Medium-term (3-6 Months)

Community Features:

  • Share your best prompts
  • Leaderboards for learning progress
  • User-submitted mentor personas

Enterprise Features:

  • Team accounts with usage analytics
  • Custom model endpoints
  • Private deployment options

Multi-language Support: Hindi, Tamil, Telugu for broader accessibility in India

Long-term Vision

  • Free and Open for All: Keep the platform completely free and open-source, allowing anyone to learn, adapt, and build upon it
  • Enterprise Integration: Enable companies to integrate Eklavya into their internal education portals with custom content and domain-specific mentors
  • Partner with AI Providers: Collaborate with Anthropic to integrate into their official learning tutorial paths and developer onboarding
  • Education Platform Partnerships: Provide white-label solutions for education companies to build custom AI learning experiences on top of Eklavya
  • Become the “Khan Academy for AI Education”: Free, high-quality, interactive AI learning for everyone
  • Expand to Web3, Security, DevOps: Virtual mentors for other critical tech domains
  • Research Platform: Enable academics to study prompt engineering patterns and model performance at scale

The goal:

Remove the fear of AI by helping people understand not just the “how” but the “why.” AI should enhance thinking, not replace it. Eklavya empowers everyone — developers, tech leads, architects, project managers, directors, students, and non-technical professionals — to upskill and confidently use AI as a tool for growth, not just copy-paste automation.

Make world-class AI education accessible to anyone, anywhere, for free — because the cost of not learning AI is far greater than the cost of learning it.

Built with Next.js 14 and Claude Sonnet 4.5

Built With

  • anthropic-claude-api-(sonnet-4.5)
  • deepseek
  • flash)
  • framer-motion
  • gemma)
  • git
  • google-gemini-api-(pro
  • gpt-4o
  • gpt-4o-mini)
  • grok-(x.ai)
  • groq-api-(llama
  • javascript
  • meta-llama
  • mistral-ai
  • mixtral
  • monaco-editor
  • next.js-14
  • node.js
  • ollama
  • openai-api-(gpt-4
  • perplexity-api
  • prism.js
  • react-18
  • react-markdown
  • recharts
  • sarvam-ai
  • tailwind-css
  • typescript
  • vercel
Share this project:

Updates