Rahul Ligma

Rahul Ligma is the public alias of Rahul Sonwalkar, founder and CEO of Julius AI in its 2023–2024 interviews and profiles. His work centers on turning natural-language requests into executable code, then using the results to guide further analysis. He argues that AI programming should reach people who have data and questions but lack coding expertise—a broader audience than professional programmers.[1][4][3]

In a Chroma interview recorded May 25, 2024, Sonwalkar described Julius as an execution loop: translate an English request into code, run it, return the output to the model, and let it decide whether to continue or try another approach. Each user received a separate sandbox with uploaded files, CPU resources and memory. He said much of the early team's effort went into securing and operating those environments and building the interface.[4]

The team moved beyond a general Python execution tool by using error telemetry to identify recurring failures and introduce specialized repair tools. Sonwalkar's example was a revenue column mixing numbers with values containing dollar signs: a preview of a few rows could miss the inconsistency. After a type error, Julius could supply instructions to inspect the affected column and remove unrelated tools from the model's context, focusing it on that repair.[4]

Sonwalkar recalled that file-import errors rose after the introduction of GPT-4 Turbo because generated paths incorrectly included /mnt/data. Prompt instructions reduced the problem without eliminating it. The team ultimately detected and stripped the unwanted prefix from the output string—a deterministic fix around an unreliable model behavior.[4]

User requests for R support changed another initial assumption. Sonwalkar said nobody on the team knew R, and they first told users that Python could do the same work. Users needed R because their existing work and collaborators used it. Adding R kernels expanded the dependencies, execution environments and specialized tools the team had to maintain, prompting investment in evaluations. These included end-to-end checks that an uploaded dataset could produce a useful visualization and checks that packages remained compatible with the models generating code.[4]

By that May 2024 interview, Sonwalkar described optional star ratings after every third AI message and smaller models, including Claude Haiku, used for classification tasks such as identifying error types. Model calls also powered ordinary interface actions: the export button asked the model to write and execute export code, while a GPT-4 graph editor derived adjustable JSON parameters from the code that generated a chart.[4]

In his November 2023 DevDay interview, Sonwalkar singled out speed and multiple function calling as especially consequential improvements. He reported that answers arriving a few seconds faster helped users progress from their first message to a second; his conversion measure concerned conversational follow-up. He also proposed requesting generated code and a documentation lookup together, so a changed Python API could be repaired without waiting for another model call. At the time, Julius looked up documentation after an error occurred.[1]

Sonwalkar described the team's addition of FFmpeg support as an example of capabilities emerging from tool access. Users requested video-to-GIF conversions, video summaries and meme edits without the team programming each task separately. Julius also generated React and HTML components; he envisioned adding a screenshot feedback loop to improve generated pages, presenting that as a possibility rather than an existing feature.[1]

Asked about competition from OpenAI, Sonwalkar said Julius had launched after Code Interpreter and still attracted 100,000 users. Interviewer swyx qualified his suggestion that this demonstrated preference: people might have used or simply tried Julius. Sonwalkar nevertheless urged founders to keep building, pointing to the effort required to make individual capabilities work well: “so many people give up before even getting started.”[1]

By May 2024, Sonwalkar said the team had simplified first use after observing that users bypassed explanatory interface elements and went straight to the chat input. Additional features could appear selectively within the conversation. His model-selection principle was similarly practical: “Our users really care about what problem we solve for them.” He said the team would use external models or train its own according to what best solved that problem.[4]

Maxwell Meyer's October 2024 profile reported that Sonwalkar had personally answered 5,000 support tickets in the year since Julius launched. Sonwalkar said scientists' enthusiasm had surprised the team. His larger ambition was for doctors, accountants, scientists and other domain experts to build software tailored to their own problems, with AI acting as a programmer in their pocket.[3]

Before Julius, Sonwalkar's college projects, hackathon relationships and engineering work supplied experience building software; the Rahul Ligma persona emerged separately from a 2022 prank.[3][6]

Sonwalkar was finishing a computer science degree at the University of Texas at Dallas in 2020, having received a scholarship. He recalled finding community and internship referrals through hackathons: “I met people who were building things just because they could.” His Y Combinator founder biography identifies work on Uber's rider pricing team and Facebook's misinformation team, and says he built the college project Waterview and grew it to 17,000 developers. Julius belonged to Y Combinator's Summer 2022 batch.[3][6]

In 2022, Sonwalkar and his friend Daniel posed outside Twitter's San Francisco headquarters as employees fired by Elon Musk, carrying boxes and giving the names Rahul Ligma and Daniel Johnson. Reporters accepted the story before discovering the prank. Sonwalkar later told Meyer: “We didn’t expect it to go viral like it did, we just thought maybe some of our friends would see it and find it funny.” Rahul Ligma was an invented employee persona, not a record of Sonwalkar working at Twitter.[3]