Joe Reeve

Joe Reeve is a software engineer who led AI R&D efforts at Amplitude in 2023 and works on ElevenLabs’ Growth team helping developers use its audio models. His work at Amplitude included turning natural-language questions into charts, diagnosing failures across multiple model calls, and using users’ subsequent edits to understand what generated answers got wrong.[1][3]

Reeve was an Engineering Manager at Amplitude when he appeared alongside cofounder and Chief Architect Jeffrey Wang on Latent Space, in an interview published June 8, 2023. The conversation was recorded during Amplitude’s June 3–4 AI x Product hackathon. Reeve described his role as running the company’s AI R&D efforts.[1][2]

His team’s question-to-chart feature let users ask about their data and receive a chart, with an early rollout to AI design partners. Hallucinations and failures to recall information within the context window led the team to split generation into seven smaller questions, including which events to use, how to display the result, and what to name the chart. The output targeted Amplitude’s custom query engine rather than SQL. Traditional ML narrowed the fields passed to the LLM; Reeve said the chart-generation calls then used GPT-3.5.[1]

Decomposition created another problem: a single user request could fail at several different steps, making one overall success metric insufficient. Reeve described instrumenting both individual inferences and groups of inferences with Amplitude to track failures, causes, and retries. After finding LangChain restrictive, the team built a small TypeScript wrapper around an inference transaction, allowing developers to write ordinary asynchronous code while monitoring the request and its constituent model calls.[1]

Reeve also reported using Postgres, followed by pgvector, for embedding search, while stressing that rollout and traffic were limited. Wang clarified that the vectors represented event taxonomies rather than individual users or behavioral events. Wang supplied the specific account of validating event selection with a general model before moving to an embedding-based approach for lower cost and latency.[1]

Reeve advocated evaluating generated content through whether people used, shared, or discarded it, alongside explicit ratings. Amplitude’s existing chart instrumentation let the team link parameter edits back to a generation: removing an event or changing a metric exposed which choice the user corrected. He described prompt A/B testing as a planned next step in June 2023, rather than an implemented system.[1]

He argued that users should understand what a model is doing and why, and be able to modify its output, stop it, or retry while it runs. His stated takeaway was: “So yeah, get machines helping humans, not the other way around.”[1]

In August 2023 commentary attributed to Reeve, he identified descriptive code naming and repetitive API “glue code” as useful areas for coding assistants. For personalization, he emphasized data governance and coordination between engineering and product teams: deciding where recommendations belong also determines which data must be collected and how intelligence can fit into the interface.[5]

Reeve authored “Talk to a Statue”, published February 18, 2026 and updated August 31. The tutorial presents a mobile web application that photographs an artwork, uses OpenAI to identify its characters and research its history, generates distinct voices through ElevenLabs, and starts a real-time conversation over WebRTC. The voices are synthetic interpretations created from descriptions, without voice samples or cloning.[4]

The implementation generates voice previews, saves selected voices, and configures a multi-voice agent whose responses can switch between characters. A server issues the conversation token, while the client’s useConversation hook handles audio capture, streaming, voice activity detection, and playback. Reeve’s article reports successfully generating the app from one prompt in Cursor with Claude Opus 4.5 (high), starting from an empty NextJS project.[4]

Reeve’s background also includes software engineering and collaborative tools. In the 2023 interview, he recalled becoming a professional software engineer at 16 and quitting college. ElevenLabs’ biography credits him with serving as CTO of the Coronavirus Tech Handbook and building a real-time collaborative editing platform used by thousands of contributors. The August 2023 Telco Magazine biography also identifies him as the founder of Gived, an R&D design and development agency.[1][3][5]