Every project was a discovery problem. Here's what we found.

Two of Sri Lanka's official languages — spoken by 22 million people — had no production-grade optical character recognition. Millions of documents (legal, medical, governmental) existed only on paper. Every other OCR provider had looked at the complexity and walked away. No training data. No existing models. No research to follow.

The technical challenge wasn't the hardest part. The linguistic challenge was. Sinhala has compound characters that change shape based on context. Tamil has stacked consonant clusters. No amount of model training would work without someone who understood how these languages actually work — not just how they look. This wasn't a problem you could solve from Silicon Valley.

Custom training data built from scratch. Linguistic rules encoded alongside machine learning models. A hybrid approach that used AI where AI excels (pattern recognition) and human knowledge where machines fail (contextual language understanding). Production-deployed. Running live.

Sri Lankan lawyers needed to research legal precedent. There was no searchable database. No API. No structured data. Legal knowledge existed in bound volumes and the memories of senior practitioners. Young lawyers without access to large firm libraries were at a structural disadvantage. You couldn't optimise a workflow that didn't exist.

This wasn't a technology problem with a technology solution. It was an access-to-justice problem that technology could address — but only if built by people who understood how Sri Lankan law actually works. The discovery was understanding that the legal system needed a new capability, not a faster version of an existing one.

AskLex.law. An AI-powered legal research platform built specifically for Sri Lankan law. We processed, structured, and indexed legal documents that had never been digitised. Built search and retrieval that understands legal concepts, not just keywords.

A major financial institution processing thousands of documents daily. The existing system was fragile, the compliance requirements were non-negotiable, and the previous vendor had created a dependency that made switching feel impossible.

The client didn't need a new system. They needed a system built by people who understood that in financial services, the compliance architecture is more important than the feature list. And they needed a team that could build without creating another dependency.

Full platform rebuild on AWS. Zero-downtime migration from the legacy system. Document processing pipeline with AI-powered classification and extraction. Built to financial services compliance standards from day one — not retrofitted afterwards. Standard architecture throughout — the client's next team could take over without us.