Sid Bendre

Sid Bendre, also known as Sidhant Bendre, co-founded Oleve, the consumer AI company behind Quizard AI and Unstuck AI. His work includes building shared infrastructure and agents that help the team research markets, source creators and reuse operating tools across products. In a 2025 interview with Nathan Latka, he reported that Oleve had reached $6 million in annual recurring revenue with six people.[2][4]

Bendre started Oleve’s platform organization alongside two other functions: product engineering, run by co-founder Ashraf, and growth and marketing, run by Ben and Mike. He described platform work as a “role of empathy”: understanding processes throughout the company and building tools to augment, scale or eliminate their steps. He explicitly said he was not the direct owner of Oleve’s App Store experiments.[4]

One concrete result was a reusable app blueprint containing accumulated coding patterns, experimentation infrastructure and notification tooling. Bendre explained that these systems were available to product engineers from the start, with common tools allowing someone to move between products without learning a different operating setup each time.[4]

To extend that platform into growth operations, Bendre said he spent time observing the marketing team’s processes and writing code to automate them. He described agents under development for creator sourcing and market research, including internal tools that studied App Store categories and app performance to identify opportunities suited to Oleve’s existing growth playbooks.[4]

Bendre’s ambition was to make Oleve a company where people exercised taste while agents handled much of the operation. His imagined interface had a “find app to build” button and a “go viral” button; these described a desired future rather than completed automation. In the same interview, he said Unstuck was probably Oleve’s last consumer education product and that a non-edtech consumer app was in development. He argued for withholding early opportunities and tactics from public view because competitors copying them could erode their value.[4]

Unstuck’s learning agent generates study notes from lecture recordings, video summaries, flashcards and quizzes. In an undated Raindrop case study, which identifies him as CTO, Bendre described monitoring agent changes as part of the product’s user experience. He said daily issue notifications influenced his priorities, including problems such as incomplete lecture notes and missing math explanations.[3]

After the team shipped a summarization change, Bendre said it filtered Raindrop’s User Frustration feedback by the summarization tool to assess whether the fix helped. Feedback also shaped feature decisions: “It directly led to us launching flashcards and quizzes!” he said.[3]

Before this platform work, Bendre joined Oleve after its first product had launched. He recalled that Mike, his friend and co-founder, asked him to help scale the AI work. Bendre had expected to go to Palantir to figure out what to build next, but said the opportunity to build with friends made joining an easier decision. His earlier technical work included DroneFormer, a language-controlled drone prototype.[4][6]

Bendre built DroneFormer with Brian Wu, Michael Tang and Ioana Marinescu. The TreeHacks 2023 submission lists the team as a Moonshot Prize winner and attributes prompting, drone control, camera-stream analytics and computer-vision decision-making to Bendre.[6][5]

The prototype used in-context learning with the OpenAI GPT-3 API to translate natural-language instructions into atomic DJI Tello actions. It combined DJITelloPy movement primitives with custom object-detection and visual-query actions using components including YOLOv5 and OpenCV. The team’s README describes a sequence in which the drone surveyed its surroundings, located the requested object, executed the task and landed; if it could not identify the object or task, it entered an abort mode and landed.[5]

The team reported trying BLIP and GLIP before selecting YOLOv5 for its balance of latency and image understanding. Limited onboard computing required a laptop between the drone and cloud inference, creating a three-tier system with frequent connectivity problems. Short battery life and a confined indoor testing environment also limited the prototype.[5][6]