I Built a Tamil Crossword Game for My Mom (And Had to Fight AI to Make It Work)

2026-02-16 · Source: Naturallanguageprocessing on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Gaming & Interactive Media · Depth: Intermediate, medium

Summary

A developer created "Tamil Kattam," a bilingual Tamil-English crossword game, inspired by his mother's Sunday newspaper ritual. Initially attempting to use AI tools like Blackbox and Supernova in 2022, the developer encountered significant challenges due to AI's inability to handle complex Tamil grapheme logic and grid intersections. The project was revived using a multi-AI stack (ChatGPT as architect, Gemini as critic, Antigravity for deterministic code) and an adversarial prompting approach to force AI models to provide accurate, rather than merely plausible, outputs. Key technical hurdles included implementing Grapheme Clustering and Unicode Normalization for Tamil characters and using the AIX Intent Tool to spoon-feed strict schema definitions to prevent grid hallucinations. The game features a drag-and-drop interface to bypass keyboard issues and a Glassmorphism UI, scaling to 100 levels using the AIX framework for pipeline generation.

Key takeaway

For Machine Learning Engineers developing applications involving non-Latin scripts or complex linguistic rules, you should adopt an adversarial approach to AI prompting. Do not rely on a single LLM for correctness; instead, use multiple models to critique each other's outputs and employ tools like AIX Intent to enforce strict data schemas. This strategy helps overcome AI's inherent biases towards Latin scripts and prevents hallucinations, ensuring robust and accurate application logic.

Key insights

Adversarial prompting and strict schema definition are crucial for overcoming AI limitations in complex, non-Latin language tasks.

Principles

• AI optimizes for plausibility, not correctness.
• Explicitly define constraints for AI outputs.
• Traditions can adapt to new digital mediums.

Method

Employ a multi-AI stack with one model acting as an architect and another as a critic. Use an adversarial prompting strategy to refine outputs. For complex logic, spoon-feed strict schemas via tools like AIX Intent.

In practice

• Use `Intl.Segmenter` for grapheme clustering.
• Implement Unicode Normalization for character handling.
• Consider drag-and-drop for complex script input.

Topics

• AI-assisted Game Development
• Multilingual NLP
• Tamil Grapheme Processing
• LLM Prompt Engineering
• AIX Framework

Best for: Machine Learning Engineer, Software Engineer, Prompt Engineer

Related on AIssential

• Engineering a Tamil Crossword: Grapheme Clusters, Auto-Cropping, and Offline-First Architecture — Developing Indic language apps requires specialized Unicode handling, dynamic layout algorithms, and robust offline capabilities.
• Comparing LLM and Fine-Tuned Model Performance on NVDRS Circumstance Extraction with Varying Prompt Complexity — LLMs, guided by a context-aware "Complexity Score" for prompt selection, significantly outperform fine-tuned models on inferentially complex, low-prevalence data extraction.
• Linguistic Issues in Computational Linguistics for Non-English Languages — English-centric LLM design creates systematic linguistic weaknesses for non-English languages across multiple levels.
• PromptMN: Pseudo Prompting Language — PromptMN structures natural language prompts with typed directives, enhancing clarity and reliability in human-AI interaction.
• I Built a System That Detects AI Manipulation Attacks — Here Is What the Results Revealed — Deep learning models like BERT significantly outperform classical NLP in detecting sophisticated prompt injection attacks.
• Code, Chaos, and the Invisible Logic of LLMs — LLMs predict words probabilistically, not with human intention, leading to potential misunderstandings in code generation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Naturallanguageprocessing on Medium.