DiliTrust

We follow an iterative development process with continuous production deployments and short feature delivery cycles. Each team is organized according to its own constraints to remain efficient and aligned with this approach. Teams are free to adapt or challenge their rituals at each retrospective. We don’t change what works — instead, we strive to objectively assess our effectiveness through measurable indicators, and we continuously improve as a team.

We’ve built a global infrastructure to vectorize and index content of all types into specialized databases and LLM-based services. These are built on open-source models fine-tuned with domain-specific data to interpret and generate text (we do not use services like ChatGPT, for client data confidentiality reasons). We continuously evaluate LLM quality, train models, and benchmark different solutions to solve business-critical challenges such as semantic search, information extraction, contract negotiation, and drafting.

We’ve unified authentication across all applications through a centralized service distributed across multiple regions. It supports all major protocols and methods required by our clients: SAML, OpenIDC, Google, Azure, simple login, and TOTP-based 2FA. To comply with GDPR, no personal data is stored — only hashed values. Application authentication follows an OAuth2 flow, and backend services across our data centers securely update authentication data via gRPC interfaces.