dau is an integrated hardware/software platform for accelerating analytical queries. It maps SQL-style operators onto a reconfigurable, tile-based dataflow engine on an FPGA, streams your data through it as Arrow record batches, and drives the whole thing from a familiar dataframe API.
A library of query operator tiles — filter, map, sort, partition, grouped aggregate, and join — wired together on an on-chip network to execute your query graph directly in hardware
Data streams through the operator tiles as Arrow record batches over PCIe, keeping the pipeline saturated for high-throughput, latency-sensitive workloads
Drive dau straight from lazy Polars — the engine is aware of which operator tiles are on the fabric and schedules each query onto them, reconfiguring the dataflow to fit and falling back to the CPU for anything it can't accelerate. Targets FPGAs from hobbyist boards to enterprise hardware
dau is an early-stage project under active development, and the design is still evolving
Reach out via email, or follow along on our GitHub.