Jeffrey Pinner

Jeff Pinner is chief technology officer of Exa, where he works on search infrastructure for AI agents. His career spans contributions to HTTP/2, marketplace engineering at Lyft, and technology leadership at Robinhood. At Exa, he has highlighted historical web search as a way to backtest models and make evaluations reproducible.

From spacecraft systems to internet protocols

Jeff’s early career included systems engineering at NASA. At Twitter, he worked on the protocols applications use to exchange information across the internet. In 2012, he submitted Twitter’s recommendations for HTTP/2, drawing on the company’s deployment of SPDY, a predecessor protocol.

The recommendations made specific engineering judgments. Multiplexing—carrying multiple requests over one connection—could reduce connection overhead and let responses arrive without waiting for earlier requests to finish. Transport encryption should be mandatory, protecting users even when application developers overlooked security. Header compression had demonstrated performance benefits, while the extra complexity of SPDY’s compression dictionary deserved closer scrutiny. Jeff was representing Twitter’s position, grounded in the company’s operational experience.

He also coauthored HPACK with Bill Gallagher, an open-source implementation of HTTP/2 header compression. The library supplies encoders and decoders for shrinking the metadata sent with requests and responses. The original HTTP/2 specification, published in 2015, credits Jeff among its substantial editorial contributors.

Infrastructure and marketplace leadership at Lyft

Jeff spent nearly a decade at Lyft, helping scale its engineering infrastructure and rideshare service, leading its marketplace organization, and eventually becoming CTO. His responsibilities expanded from the systems supporting the product to the marketplace connecting riders and drivers. He was serving as CTO when Lyft reorganized its leadership in April 2023.

During his Lyft years, Jeff continued working on network design. With Roberto Peon, he coauthored a 2018 QUIC abstractions proposal. Their draft separated packet transport, connections, streams, and HTTP into layers with distinct responsibilities. One concrete problem involved related video streams: separating independently useful components could improve delivery, but those components still needed to reach the same application endpoint. The proposal explored grouping streams to preserve that relationship. It remained an Internet-Draft rather than becoming an adopted standard.

After Lyft, Jeff became a distinguished engineer at Cruise. He then joined Robinhood as CTO in August 2024, moving from transportation systems into financial software, before subsequently joining Exa.

Searching the web as it was

Exa builds and operates web-scale indexes and trains neural retrieval models. Its search infrastructure determines what information agents can retrieve, including which historical version of a webpage they can access.

Jeff called Exa Snapshot his first product launch since joining Exa. He emphasized historical web data as a resource for backtesting models, training on the pre-AI web, and creating reproducible evaluations.

Snapshot searches historical webpage versions. Filtering current pages by publication date does not restore their earlier contents: a page may have changed to include an answer that did not exist when a test was written. Searching historical versions lets researchers evaluate an agent against an earlier information environment and financial modelers test predictions using information available at the time. Jeff’s announcement identifies the uses he considers important without establishing individual authorship of this team product.