Raunak

Big milestone for Reducto: we’ve raised a $75M Series B led by a16z, bringing our total funding to $108M. Since our launch, we’ve been obsessed with pushing the limits of what’s possible when AI meets the real-world messiness of documents. We’ve now processed 1B+ pages, nearly 6x the pages since our Series A less than 6 months ago. We’re doubling down on model research, developer tooling, and the reliability that production AI demands. We'll be announcing a series of product improvements that will extend our capabilities even further, helping our customers get even closer to perfect accuracy across their most challenging docs. A big thank you to everyone who helped get us here. But we're just getting started - if you're interested in what we're doing, we’re hiring across all functions! reducto.ai/careers

@adelwu_ banger event

how did 4 ppl thread a photo of themselves at my event LOL

have been working with a lot of our customers on debugging workflows where agents work with document content and captured some of the learnings here!

you can tell claude planned this because it was scoped for four weeks but actually completed in under an hour

A little biased, but I would try @reductoai first - blueprints tend to be pretty out of distribution though it really depends on what you're looking for. If you're doing something really custom like counting objects from the legend in the blueprint diagrams you may want to build your own custom stuff on top of Gemini 3.1 Pro which is fairly good at bounding boxes.

@raunakdoesdev @thenanyu What model do you recommend? I'm looking to parse construction blueprints.

Is it better to just render PDFs as images and OCR them rather than read directly then?

@theo submitted! github.com/raunakdoesdev/…

I would like to purchase a handful of code problems that modern LLMs can’t solve. Requirements: - programmatically verifiable (can be tested without human interaction) - “before” state (repo before the commit that implements the solution) - example code that actually solves the problem I am willing to pay up to $500 per problem that I can easily test locally and confirm current models (gpt-5.3-codex, opus 4.6) are unable to solve. If you can’t tell, I’m running out of “too hard for LLM” code tasks 🙃🙃🙃

@benhylak holy shit this is genius 🤯

today, we're introducing self diagnostics: the first ever way for agents to proactively self-report issues they encounter. welcome to the future of agent observability.

@SarahChieng yes! would love an intro

if anyone's hiring: got a great Harvard friend looking for the next thing :) > Harvard Computer Science > Strong UX design, product, and CS background > Prev. Microsoft, Snowflake, startups They're a quick leaner with a background across product engineering, UX Research/Engineering, and design. If anyone has any opportunities, they'll be following this post. (please comment, do not DM me)

this is amazing! unique approach to trace exploration

@charlieholtz @iannuttall @conductor_build really looking forward to this!

@iannuttall @conductor_build Sorry about this, we just fixed a bunch of perf issues in 0.35.3 but clearly more to improve on. Should have more this week!

Does @conductor_build grind to a halt after ~1h+ usage for anyone else? Seems to get so slow and laggy for me and really breaks my flow :(

@Winterrose supreme dumplings is underrated

HK Lounge Bistro is my favorite XLB restaurant in SF if you know one that can unseat it, tell me. I dare you.

Introducing Reducto for Startups. If you’re building on top of complex, document-heavy workflows, we’ve put together programs to help you run document extraction reliably in production, with credits, startup-friendly pricing, and hands-on support. Designed to help teams move from prototype to scale without managing document infrastructure in-house. Apply at: YC → reducto.ai/yc Startups → reducto.ai/startups

pretty incredible turn around time from the @circlebackai team on this actually. bullish!

any good alternatives to @circlebackai? has gotten too buggy/slow lately. want: 1. good developer experience (webhooks, api, etc.) 2. integrations with attio, slack, etc. - ideally can build basic automations within the product. 3. actually joins the call (no secret recording)

@rag_pil @circlebackai it's actually mainly #3 - having video recordings is helpful for a lot of meetings (screen shares etc.) and for certain important meetings like sales/cs don't want to rely on users remembering to toggle on granola for it work (should be server side/automatic)

@raunakdoesdev @circlebackai other than #3 what was ur reason for not using granola? granola mcp + claude should go hard here

@darrenjr would be sick if y’all could do this on pylon tickets as well

whenever i get a new sentry error, an instance of claude code with opus 4.6 in a vm spins up automatically to fix it

@Winterrose el norteno truck in soma is the only right answer here

El Farolito in North Beach fuckin mogs La Taqueria. both are still mid AF compared to the burrito trucks of Pasco WA. BUT El Farolito is S-tier for SF.

@agupta @charlieholtz @conductor_build can confirm conductor is the shit

@charlieholtz @conductor_build working my way through the apps :D

okay i will admit the codex desktop app is WAY nicer than the claude code app

We recently released a paper on Activation Oracles (AOs), a technique for training LLMs to explain their own neural activations in natural language. We piloted a variant of AOs during the Claude Opus 4.6 alignment audit. We thought they were surprisingly useful! 🧵