The Lab
Research, experiments, and engineering at the frontier.
Everything we learn here ships into production.
RAG Pipeline v3
Third-generation retrieval-augmented generation pipeline. Hybrid search combining dense vector + BM25 sparse retrieval with re-ranking. Context window management for 128K+ token documents.
Voice-to-Structure
Real-time voice transcription to structured data pipeline. Whisper + Claude for converting spoken input into JSON schemas. Used in clinical documentation and field reporting.
Agentic Workflows
Multi-agent orchestration framework for complex business processes. Task decomposition, tool use, and self-correction loops. Production-tested across invoice processing and customer support.
Edge Inference
Running quantized models at the edge for latency-critical applications. Exploring ONNX Runtime and WebGPU for browser-native inference without server roundtrips.
Document Intelligence
Production document understanding pipeline. Layout-aware parsing, table extraction, and semantic chunking for invoices, contracts, and medical records.
Open Source
We contribute back. Tools we build for ourselves become tools everyone can use.
View on GitHub →