Case Study

Lila

Natural-language BI across 95+ languages.

AI-powered business intelligence platform that lets non-technical users query databases by typing in plain language. Architecting the LLM layer at AALA, wiring OpenAI, Claude, and Mistral together with caching, prompt optimization, and fallback logic for when the model gets it wrong.

What it does

A user asks 'show me last quarter's churn by region' in English, Urdu, Vietnamese, or anything else. Lila translates intent into SQL, runs it against the customer's warehouse, and returns the answer with the query exposed so analysts can verify.

Where AI helps and where it doesn't

The model writes the SQL. Everything around the model is the actual product: schema introspection, query validation, cost controls, caching of expensive prompts, and a fallback path for when the LLM hallucinates a column that doesn't exist. The hard part is never the API call.

Tech Stack

OpenAI Claude Mistral Next.js PostgreSQL Vector DB