why right now feels like the start of everything
it’s been a while since i wrote one of these. a proper blog, a long one, not just an update or a snippet. finals ended. practicals are over. the semester is about to start again, and somehow it feels like a pause and acceleration happening at the same time. that weird moment where nothing seems urgent, yet everything feels like it’s moving faster than it has in months.
sagea has been quietly, steadily, relentlessly moving forward. and the pace of the quiet acceleration is starting to be noticeable. the kind of acceleration that isn’t flashy or hyped, but that hits you in the chest when you pause and look at the months behind you. models shipped, products released, community engagement building, traction slowly emerging, and yes, the occasional investor noticing, asking the right questions, sending the right emails. all of it adding up to a feeling that this is the start of something bigger than the sum of its parts.
the most obvious milestone has been the open-sourcing of the SAGE reasoning family. 3b, 8b, and 14b hybrid reasoning models. all built with iterated distillation and amplification, layered on top of the inverse reasoning head; our metacognitive novelty at sagea. the response has been phenomenal. the community is engaging. developers are experimenting. students are implementing variants. discussions are sprouting across forums and chats. people are building with the models in ways that we hadn’t anticipated. there’s a ripple effect that feels almost tangible. and it’s validating, in a quiet but deep way.
it’s not just about attention or recognition. it’s about seeing the impact of the work. seeing someone post a pull request that actually improves reasoning behavior. seeing a student write a small project that chains inverse reasoning to solve a problem. seeing someone reference SAGE in a discussion about how to approach a complex reasoning task. the feeling is different from metrics or downloads. it’s qualitative. it’s the kind of validation that makes the late nights, the debugging marathons, the countless experiments, all worth it. it’s addictive in a way that’s hard to describe.
beyond open-sourcing the models, we’ve been experimenting with agentic products. products that don’t just respond, but act. inscribe was the first real test. a paper-to-code agent. ujjwal led development. and the reception has been better than expected. the idea is deceptively simple: take academic papers, turn them into executable code. in reality, it’s insanely complex. papers are inconsistent, experimental setups are ambiguous, math is dense. and yet, inscribe works. in labs, in small projects, in personal experiments. developers are testing it. students are integrating it into workflows. early adopters are remixing it. and the feedback isn’t just positive. it’s thoughtful. people are thinking about what’s possible when they can take research and execute it automatically.
on my side, firoj and i have been working on something parallel and deeply exciting: algor, our coding CLI agent. i’m leading development. firoj is focused on evaluation and benchmarking. early results are promising. algor reasons, it produces code reliably, it executes tasks with a sense of discipline that feels almost alive. the first time i saw it solve a problem that had stumped me for hours, i paused and realized we were doing something that felt fundamentally different from anything we’d shipped before. it’s still early days. we have a lot to refine. but the promise is there. it’s rare to feel a product shift from concept to reliable execution in such a short time. it’s motivating, addictive, and humbling all at once.
behind the scenes, the SAGE MoE variants have been quietly, slowly, astonishing us. models twice their size are being outperformed. it’s one of those moments where the architecture choices, the reasoning heads, the routing, the design decisions; every single thing we debated for months, suddenly feels like it paid off. we’re still in R&D. every benchmark is a lesson. every test run is a conversation between what we expect and what actually happens. and yet, the trajectory is visible. a 40b MoE that can outperform models far larger than itself is not just a number on a spreadsheet. it’s a signal. it’s a hint at what’s possible when design, reasoning, and compute efficiency intersect.
popularity has been slow but real. SAGEA is becoming a name. the models are being used, talked about, referenced. not flashy headlines yet. not “everyone knows our name” level. but enough visibility to make conversations with potential investors meaningful. VCs are noticing. quietly. casually. conversations turning into meetings. meetings turning into deeper discussions. traction isn’t hype. attention isn’t vanity. they’re signals. and we’re in a rare position where we have both. consistent output, real products, emerging credibility; the math is working, the compounding is visible.
team culture has been evolving in parallel. three main people, a handful of contributors. tiny, but growing in spirit. the first open-source release shifted the dynamic. late-night debugging is now paired with celebrations. the first time algor solved a puzzle we couldn’t crack, ujjwal laughed at 2am. small moments, seemingly inconsequential, but they compound into a morale that sustains you when models fail, when code breaks, when APIs lag, when no one replies to your outreach. curiosity is normalized. obsession is encouraged. chaos is structured. it’s messy, but it works.
random bops, small external events, have been keeping us grounded. we went to a local meetup last week. students were trying to replicate SAGE outputs. one of them said, “you’re literally years ahead of what I expected.” it’s humbling. it’s a reminder that we’re not just building for ourselves. that we’re impacting the ecosystem, the next generation of builders. tiny validation, but it carries weight.
other small wins: inscribe generating a working function directly from a paper without edits. algor producing a solution that passed all unit tests on the first try. SAGE MoE benchmark numbers exceeding expectations. moments like these accumulate. they’re small, but they compound into confidence, into momentum, into belief in what we’re building.
on a personal level, it’s surreal to be here. finals over, semester about to start, projects alive, traction building, models behaving, community responding. it’s exhausting and thrilling at the same time. the kind of exhaustion that feels like it’s productive. that feels like it’s building something permanent. it’s rare to feel calm and accelerated simultaneously. that’s the feeling i want to capture today.
looking forward, the next months are going to be a blend of iteration, refinement, and expansion. shipping algor, refining SAGE MoEs, releasing more agentic applications, maintaining community engagement, continuing investor discussions. each of these elements matters. every call, PR, line of code, benchmark compounds. the work is deliberate. the growth is deliberate. nothing is accidental. momentum isn’t hype. it’s consistent action that scales.
and yet, even as everything feels deliberate, there’s room for chaos. small experiments, tangents, side projects. testing ideas that might fail. that’s part of building a real product. that’s part of building a team culture that doesn’t fear failure but embraces it as learning. the balance between structure and chaos, output and iteration, curiosity and discipline; that balance is what makes the work feel alive.
outside models and code, the ecosystem is evolving. the community around SAGE is growing. students are building, contributing, remixing. early adopters are sharing insights. the feedback loop is real. the social layer, the human layer, is amplifying the tech layer. traction is no longer just internal. it’s emergent. organic. and that’s rare. that’s valuable.
life continues. semester begins. meetings, classes, labs, late nights. but there’s clarity now. the models, the products, the community, the traction, the team, even the timing; it all aligns. sagea is starting to feel inevitable. small, deliberate, almost invisible to outsiders, but unstoppable to those inside. the quiet acceleration is happening. and being here to feel it, to observe it, to participate in it, is something rare.
and so i write this today. not hype. not overconfidence. just reflection, observation, gratitude, and quiet excitement. about models that think. agents that act. a team that builds. a community that responds. traction that grows. and the compounding effect of all of it converging at the same time.
this is momentum. this is alignment. this is the slow roar that precedes something larger than any of us alone. and it feels rare. and it feels real.
ciao, basab
Want to get notified everytime i write stuff? Sign up to my newsletter