SAGEA is live, VORA's teased, and L1’s already open.

i’ve been heads-down on something big, and it’s time to let it loose. me and my co-founder ujjwal just announced sagea, our new company building ultra-efficient ai for voice, language, and vision. it’s made to scale fast and feel alive, like it’s not just tech but something that clicks with you. i’m dropping nexora—sagea’s where my heart’s at now. tonight, we’re teasing our vora voice model family, and we’ve already open-sourced our l1 model on hugging face and github. exams are looming, but we’re pouring everything into sagea’s foundation. work is making alot of progress too. the deep research engine is slowly picking up pace. today we discussed whether to stay with our primary provider for web-crawling or switch to the more supposedly better anthropic webresearch. don't know if we'll switch. but that's a story for another day. last week, i was wrestling with a python script that kept choking on jumbled funding data, like csvs with broken formats. i rewrote the preprocessing to handle it, and when it spotted a startup’s quiet traction, it was like flipping a switch. that kind of efficiency—doing more with less—is what sagea’s about. we want ai that’s sharp, fast, and doesn’t need a data center to run. and we seem to be on the right track with it. sagea’s our shot at that. it’s voice that sounds so real you’d swear it’s human, language that gets the vibe of a convo, and vision that picks up details others miss. VORA, our voice family, is a big piece—models that can do accents, emotions, everything, on minimal compute. we’re teasing vora tonight, showing what it can do, but the full drop’s coming later. our v1 model, the heavy hitter, is built to be powerful yet lean. but we couldn’t wait to share something, so we open-sourced l1, our lightweight model, on hugging face and github. it’s not v1’s beast mode, but it’s fast, efficient, and ready for anyone to play with. we’re hyped to see what people build with it. right now, me and ujjwal have been busy trying to figure out the infrastructure, finding the right amount of compute we need, the VPSs and all that. hopefully, we're able to launch platform access for VORA by the first week of june. until then, it's mostly going to be teasing our models, showcasing the benchmarks with our models, and all the research papers. never in my life i'd imagine i'd have an AI company like this. but things look promising, the numbers and benchmarks prove we're headed on the right track. work’s been a masterclass in scale. the vc engine deals with data at a level apparatus never touched, teaching me how to build ai that holds up under pressure. sagea needs that, but also something deeper—ai that feels human, not cold. exams are on our heads, but we’re carving out time to nail the infrastructure, test mvps, and get VORA out there. going public with sagea was a rush—ujjwal and i are just two dudes with a vision, but it’s real now. vora’s tease tonight is the next step, and l1’s open for anyone to jump in. sagea’s why i’m up late, thinking about what ai can be. it’s efficient, powerful, and alive, built for a world that’s moving fast. apparatus was a warm-up; this is the real game. we’re small—me, ujjwal, some scripts—but we’re dreaming big. anyways, VORAs tease drops tonight. more soon.. — ciao, basab