Applications shipped, infrastructure milestones, capability updates. This page is how you can tell whether we're still moving.
First client engagements live. Sovereign AI infrastructure available for professional services firms in Australian cities. Applications built on DeepSeek V4 Flash running on dedicated GPU hardware co-located in the client's city. Custom application builds, RAG-first document intelligence, and ongoing managed service.
Completed internal testing and benchmark validation for DeepSeek V4 Flash as the primary recommended model. 284B parameters, 13B active per token, 1M context window, MIT licensed. Benchmarks within a fraction of Claude Opus 4.6 on document analysis and drafting tasks. Adopted as the standard production deployment for new client builds.
Standard operating procedure for hardware procurement, configuration, and co-location arrangement at professional datacenters in any major Australian city. Lead time from signed contract to live system: 4–6 weeks for standard builds. Hardware specification confirmed: 2× NVIDIA L40S, AMD EPYC, 512GB DDR5, 2TB NVMe.
Production configuration for document indexing and retrieval: Qdrant vector database, custom chunking pipeline for legal and professional services document formats (contracts, correspondence, PDFs, Word documents), embedding pipeline with multilingual support. Indexing throughput: ~50,000 pages/hour on reference hardware.
vLLM deployment configuration finalised for serving DeepSeek V4 Flash on the L40S configuration. OpenAI-compatible endpoint, streaming support, tool-use calls confirmed working. Concurrent user testing: 3–5 simultaneous users at acceptable latency on the reference two-card configuration.