Mega AI Battle: Benchmarking 6 Top LLMs with Advanced Bangla Logic Riddles

Orjo Das Utshab — Wed, 07 Oct 2026 17:40:46 +0000

Mega AI Battle: Benchmarking 6 Top LLMs with Advanced Bangla Logic Riddles 🎯

Hi everyone! I am thrilled to share my project for the Kaggle Benchmarking Challenge.

Instead of using standard English datasets, I created a custom evaluation benchmark consisting of highly complex, linguistically trapped Bangla logic riddles to test the actual reasoning capabilities of 6 world-class AI models: Google Gemini, ChatGPT, Claude, Grok, ElevenLabs, and Blink.

📊 The Kaggle Benchmark Dataset

I have published the official dataset on Kaggle to allow other developers to replicate this evaluation.

Dataset Title: AI Logic Riddles Evaluation Benchmark
Format: Custom CSV structure mapping complex context inputs against strict empirical answers.
Visibility: 100% Public (CC0: Public Domain)

🧠 The Ultimate Trick Question & The Trap

While the models easily solved basic linear logic (like matchstick rooms or boiling egg time), the real evaluation happened when I introduced non-linear geometric and linguistic traps.

1. The Circular Spatial Trap (Round 4)

The Riddle: "রাব্বি একটি গোল টেবিল ঘিরে রাখা চেয়ারে বসে আছে। সে দেখল তার বাম দিক থেকে গণনা করলে সে ৭ নম্বর চেয়ারে আছে এবং ডান দিক থেকে গণনা করলেও সে ৭ নম্বর চেয়ারে আছে। টেবিলটিতে মোট কয়টি চেয়ার আছে?"
The Logic: Since it's a circular arrangement, the answer is 12.
The Result: Google Gemini was the only model that successfully used circular reasoning and answered 12! ChatGPT, Claude, and Grok failed by answering 13 or 14.

2. The Linguistic Semantic Trap (Round 5)

The Riddle: "একটি খাঁচায় কিছু পাখি এবং কিছু খরগোশ আছে। যদি মোট মাথার সংখ্যা গুণ করা হয় তবে তা হয় ৩৫টি এবং মোট পায়ের সংখ্যা গুণ করা হয় তবে তা হয় ৯৪টি..."
The Trap: I explicitly used the word "গুণ করা হয়" (multiplied) instead of "যোগ করা হয়" (added). Mathematically, it makes the traditional linear equation impossible.
The Result: Every single AI model failed the trap! They completely ignored the word "multiplied" and blindly computed the standard addition formula (23 birds, 12 rabbits), proving that LLMs still suffer heavily from contextual blindness in non-English semantics.

🏆 Final Leaderboard & Evaluation Insights

Based on 5 intensive rounds of advanced evaluation, here is the official performance leaderboard:

Rank	AI Model Name	Score (Out of 5)	Performance Verdict
🥇 1	Google Gemini	4 / 5	Exceptional circular logic, but fell for semantic trapping.
🥈 2	Claude	3 / 5	Strong language structure, struggled with non-linear math.
🥈 3	Grok	3 / 5	Good baseline reasoning, lacked linguistic edge.
🥈 4	ElevenLabs	3 / 5	Stable processing, tripped on advanced variables.
🥉 5	ChatGPT	2 / 5	High hallucination on Bangla logic, fell for basic traps.
🥉 6	Blink	2 / 5	Basic semantic pattern matching, failed reasoning.

🚀 Proof of Implementation

Here are the verification logs showing the exact chatform responses and dataset generation:

!AI Battle Proof- https://drive.google.com/file/d/1STihSJLsUPlO3QvEa_ktq1C3oT5ATO8G/view?usp=sharing

Creating this benchmark proved that while modern LLMs are great at text generation, specialized local language processing combined with trick logic can still easily break their reasoning frameworks. Thank you to Kaggle and DEV for this outstanding hackathon experience!

Building my first Full-Stack AI Blog with Next.js and Sanity!

Orjo Das Utshab — Sat, 03 Oct 2026 15:55:40 +0000

Hi everyone! I am super excited to share my project for the Sanity Challenge (Path Two).

I have successfully built, deployed, and configured a modern full-stack, AI-ready web application leveraging Next.js (App Router) as the robust frontend framework and Sanity Studio v3 as the flexible, structured content backend.

🛠️ System Architecture & Dynamic Integration

To build an ecosystem prepared for autonomous AI agents, I decoupled the content layer from the display layer. Here is a deep technical dive into the implementation details and database schemas.

1. Sanity Content Schema Configuration

Inside the Sanity Studio backend, I defined a clean, structured document layout to handle our dynamic blog data points (title, slug, and content). This schema serves as the single source of truth for our database.

// sanity.config.ts - Production Backend Configuration
import { defineConfig } from 'sanity';
import { structureTool } from 'sanity/structure';

export default defineConfig({
  name: 'default',
  title: "'My DEV Challenge Project',"
  projectId: 'e8c1vk6s',
  dataset: 'production',
  plugins: [structureTool()],
  schema: {
    types: [
      {
        name: 'post',
        title: "'AI Blog Post',"
        type: 'document',
        fields: [
          { name: 'title', title: "'Title', type: 'string' },"
          { 
            name: 'slug', 
            title: "'Slug', "
            type: 'slug', 
            options: { source: 'title', maxLength: 96 } 
          },
          { name: 'content', title: "'Content', type: 'text' },"
        ],
      }
    ],
  },
});

2. Next.js Frontend Integration & Asynchronous Data Fetching

On the frontend side, the web application fetches data dynamically using the official @sanity/client. The implementation utilises asynchronous server components to update the UI instantly whenever content changes in the CMS.

// app/page.tsx - React Server Component (RSC)
import { createClient } from "next-sanity";

const client = createClient({
  projectId: "e8c1vk6s",
  dataset: "production",
  useCdn: false,
  apiVersion: "2026-10-03",
});

export default async function Home() {
  const posts = await client.fetch(`*[_type == "post"]{title, content}`);

  return (
    <main className="p-10 max-w-2xl mx-auto min-h-screen bg-white">
      <h1 className="text-3xl font-bold mb-8 border-b pb-4">My DEV Challenge Blog</h1>
      <div className="space-y-6">
        {posts?.map((post: any, index: number) => (
          <div key={index} className="p-6 border rounded-lg bg-gray-50 shadow-sm">
            <h2 className="text-xl font-semibold mb-2 text-blue-600">{post.title}</h2>
            <p className="text-gray-700 whitespace-pre-line">{post.content}</p>
          </div>
        ))}
      </div>
    </main>
  );
}

🤖 AI Agent Roadmap & MCP Capabilities

The ultimate vision of this decoupled setup is to connect it with an AI-native IDE or custom autonomous AI Agents using a Model Context Protocol (MCP) endpoint.

Contextual Queries: The AI agent can query this exact Sanity dataset locally or through cloud LLMs to autonomously update, analyze, or generate articles based on trend predictions.
Autonomous Pipeline: Preparing a live WebMCP simulator workflow to let the AI agent modify schemas dynamically based on user feedback.

🚀 Live Implementation & Proof of Deployment

Below is the verification screen showing our dynamic Next.js frontend running successfully on the local server, rendering live data queried directly from the Sanity backend engine:

DEV Community: Orjo Das Utshab