Al-Optimized Themed Crossword Generation with a Multi-Agent Puzzle System on AWS for 24/7 Games

Client

24/7 Games

Location

India

Industry

Gaming & Entertainment

Services & Tech

Amazon Bedrock, AWS Lambda, Amazon API Gateway, Amazon RDS or Amazon DynamoDB

Client

24/7 Games

Location

India

Industry

Gaming & Entertainment

Services & Tech

Amazon Bedrock, AWS Lambda, Amazon API Gateway, Amazon RDS or Amazon DynamoDB

Project Overview

24/7 Games, a leading online crossword platform, wanted to improve puzzle quality beyond what its brute-force, word-list driven approach could deliver. The existing method produced playable puzzles, but limited fill rates, reduced edge-to-edge coverage, and constrained thematic richness. Avahi built a multi-agent Al system that generates themed crossword grids with optimized word placement, validates solvability, and automatically creates hints with adjustable difficulty. The result is higher-quality puzzles that maintain theme consistency and can scale to batch generation.

About The
 Customer

24/7 Games is an online gaming and entertainment platform specializing in crossword puzzles and word games. It serves a large audience of daily players by delivering puzzles across multiple difficulty levels, formats, and themes, with an emphasis on repeat engagement and fresh content.

The 
Problem

24/7 Games relied on a brute-force crossword generation approach using predetermined word lists. While functional, this method often produced less-than-ideal grids, including suboptimal fill rates, weaker word intersections, and limited edge-to-edge coverage, all of which can reduce puzzle satisfaction and replay value.

To improve content quality and scale puzzle creation, the platform needed an Al-driven system capable of generating high-quality themed puzzles with better word placement optimization. It also needed automated hint generation that could adapt to easy, medium, and hard difficulty levels, while maintaining thematic consistency and ensuring puzzles were solvable. The system had to support batch generation and multiple grid sizes without compromising quality.

Why AWS

AWS enabled 24/7 Games to implement an agentic workflow that could generate and validate puzzles at scale, while integrating cleanly with existing crossword systems through APIs. Amazon Bedrock provided access to foundation models for word generation, hint creation, and validation logic, and serverless orchestration made it practical to run multi-step puzzle generation pipelines efficiently.

With AWS managed services, the solution could store word lists, themes, and generated puzzles in a durable database layer, and expose integration endpoints through API Gateway for reliable connection to the existing crossword generation system.

Why 24/7 Games Chose Avahi

24/7 Games chose Avahi for its ability to design an end-to-end multi-agent architecture with built-in quality controls, not just content generation. Avahi delivered a system that optimizes word placement and grid coverage, validates solvability using both Al and existing verification logic, and creates difficulty-scaled hints, all whilekeeping the solution integration-ready for the customer’s current platform.

Solution

Avahi developed a multi-agent Al system designed to generate themed crossword puzzles with strong validation and quality control. The generation workflow starts with a Word Generation and Grid Placement Agent that selects candidate words and places them into the grid while maximizing fill rate, word intersections, and edge-to-edge coverage. The agent uses NLP-based logic to optimize placement and increase the number of valid intersections, producing denser, more satisfying grids.

To enforce theme quality, Avahi implemented a thematic word selection system with semantic filtering across three distinct themes. This ensures that selected words align with the puzzle’s theme and reduces off-theme filler that can dilute the player experience.

A second Validation Agent evaluates puzzle solvability and completeness. This agent integrates with the customer’s existing verification engine to confirm horizontal and vertical word validity, and it provides structured feedback back to the first agent. This feedback loop enables iterative improvements, correcting placement issues, filling gaps, and improving overall grid integrity before final output.

The system also generates hints automatically for each puzzle word and scales hint difficulty for easy, medium, and hard puzzles. It supports multiple grid sizes, including 6×6, 12×12, and 20×20, and integrates with the existing crossword generation pipeline through API endpoints. The full workflow is orchestrated using AWS Lambda, and Amazon Bedrock supports generation, validation reasoning, and hint writing. Puzzle data and word/theme assets are stored using Amazon RDS or DynamoDB, depending on the customer’s preferred persistence model.

Key Deliverables

  • Multi-agent crossword generation system with quality control and validation
  • Word generation and grid placement agent optimizing fill rate, intersections, and edge-to-edge coverage
  • Thematic word selection with semantic filtering across three themes
  • NLP-based placement optimization and intersection maximization logic
  • Validation agent ensuring solvability and completeness
  • Integration with existing verification engine for horizontal and vertical word validity
  • Feedback loop from validation agent to generation agent for iterative improvements
  • Automated hint generation with easy, medium, and hard difficulty scaling
  • Support for 6×6, 12×12, and 20×20 grid sizes
  • API integration with the existing crossword generation system

Project
 Impact

24/7 Games improved puzzle quality by moving from brute-force word-list generation to Al-optimized grid construction with validation safeguards. The new workflow increases fill rates and grid coverage, strengthens word intersections, and produces themed puzzles with difficulty-appropriate hints. It also establishes a foundation for batch generation at scale without sacrificing puzzle integrity

Metrics

  • Supported grid sizes: 6×6, 12×12, 20×20
  • Hint difficulty tiers: easy, medium, hard
  • Theme coverage: three distinct themes

Ready to Transform Your Business with AI?

Let’s explore your high-impact AI opportunities together in a complimentary session

Ready to Transform Your Business with AI?

Let’s explore your high-impact AI opportunities together in a complimentary session

Ready to Transform Your Business with AI?

Book Your Free Ignition AI Workshop

Let’s explore your high-impact AI opportunities together in a complimentary half-day session

View Our Case Studies

See how we’ve delivered measurable results for businesses like yours