The Philosophy: Intelligence belongs in the binary. I replace heavy Python runtimes with lean, GPU-accelerated Rust engines (Burn.rs | JEPA | Candle).
| Pillar | Technology Stack |
|---|---|
| Deep Learning | Burn.rs (WGPU/CUDA), JEPA, Candle, ONNX Runtime |
| Systems Core | Rust (Tokio/Axum), Java (Loom/Spring Boot), Go (gRPC) |
| Infrastructure | Nix, Docker, Kubernetes, GitHub Actions |
| Data Strategy | ClickHouse, Kafka, Vector DBs (Qdrant/Milvus) |
- The Mission: Migrate 400ms Python-based recommendation pipelines to a standalone Rust binary.
- The Build: Used JEPA for world-modeling and Candle for high-fidelity inference.
- The Impact: Achieved 45% reduction in latency and 60% faster cold-starts for millions of concurrent users.