Morphik (YC X25)

This tweet by Andrej Karpathy about Nano Banana Pro shows that image generation models can now reason and problem solve. To me, it hints toward a future where the best user experiences will intelligently weave-in different modalities in their responses to users. Memory for such user experiences will need to be multimodal too. This is exactly the future that we've built Morphik for.

A customer spent 6 months with 6 engineers building an internal retrieval system for their agents. Best accuracy? 64%, unusable for their workflow. When they handed us their eval set, they said: “If you can hit 80%, that’s a HUGE win.” First test on Morphik (YC X25) (no fine tuning, custom mods): 97% correct. Customer: “This seems too good to be true… did you guys overfit?” They spent a week creating a brand-new eval set, never shared with us. On the next call, they ran it in the background while we talked. At the end: 100% correct. Here’s the bigger point: in a world where hallucinations are front and center, accuracy isn’t just a metric, it’s the foundation of trust. - Consistent accuracy builds confidence. - Overconfident errors destroy trust quickly. - Blind evals are the fastest way to prove credibility. Curious to hear: beyond accuracy, how do you show reliability and build trust with your customers, especially with non-deterministic technology?

The “knowledge tax” is bigger than it looks. Unplanned downtime tied to human factors costs manufacturers billions every year. Add in aging teams, scattered documentation, and hours lost to “who knows this?” and the real bill is staggering. In custom manufacturing, it bites hardest: every job looks different, and the answers live across PDFs, drawings, emails, photos, even handwritten notes. The result is heroic individuals carrying institutional memory… and a system that doesn’t. The companies winning are treating knowledge as infrastructure: capture it, search it, re-use it, and let it shape how work gets done. First-pass quality improves, quoting speeds up, onboarding shortens, and weekends stay weekends. I pulled together a pragmatic breakdown of the problem + a rollout plan (pilot → prove value → expand). If you’re in manufacturing, this will feel familiar. 👉 Read: https://lnkd.in/eTiWA7nR Curious what I missed?

"We understand you're a young company and we'd like to grow with you." Heard this from a customer recently, and it reminded me why the best business relationships aren't just transactional. They chose to work with us not despite being early-stage, but because of it: → We can adapt quickly to their needs → They get direct access to our team → We're incentivized to make them wildly successful Established vendors optimize for processes. Growing companies optimize for outcomes. Sometimes betting on the underdog pays off for everyone. When was the last time you chose potential over polish in a vendor relationship?

We crushed baseline AI accuracy by 7x in our latest evaluations. Most enterprise AI pipelines fail when it comes to complex financial analysis. We've seen this repeatedly with our customers. So we built an open-source evaluation specifically for multi-hop reasoning, time-series analysis, and complex document workflows. The results? Baseline (OpenAI file search): 13.3% accuracy Custom LangChain + SOTA OCR and chunking: 66.67% accuracy Morphik (YC X25): 95.56% accuracy That's a 7x improvement over baseline and 40% better than custom solutions. Some real-world use cases where this matters: - Investment firms analyzing 10-K filings - Insurance companies processing claims with complex medical documentation - Legal teams extracting contract terms from multi-page agreements The difference between 67% and 95% accuracy? A system you can actually trust with business-critical decisions! We have benefited immensely from open-source, so we made the evaluation open-source as well! Happy to take contributions to add your results on there. Check it out: https://lnkd.in/dFp_BErw Building something that needs bulletproof document understanding? Let's chat: founders@morphik.ai. Try it out: morphik.ai

We just hit a huge milestone: 1 MILLION ingestions in Morphik (YC X25)! Just a couple months ago, we were at 500 ingestions (all us, stress-testing prod and joking about how cool it’d be to 10x). Now it’s 2000x. And the best part is seeing user stories behind the numbers. Last week, on a live call, a customer tried our self-serve platform with an “impossible” legal document: messy tables, figures, and complex layouts. Every state-of-the-art parser and layout tool had failed. One shot with Morphik, and they got the answers they needed: no hacks, no manual cleanup. Morphik is the one API to store, search, and build agents over private data, no matter how complex. We're just getting started. Would be cool to hit 10 MILLION next!!

At Morphik (YC X25), we're dedicated to building the best RAG and document-processing systems in the world. Morphik works particularly well with visual data. As a challenge, we were trying to get it to solve a Where's Waldo puzzle. This led us down the agent rabbit hole and culminated in an agentic document viewer which can navigate the document, zoom into pages, and search/compile information exactly the way a human would. This is ideal for things like analyzing blueprints, hard to parse data-sheets, or playing Where's Waldo :) In the demo below, we ask the agent to compile information across a 42 page 10Q report from NVIDIA. Test it out at https://morphik.ai ! Soon, we'll be adding features to actually annotate the documents too - imagine filing your tax forms, legal docs, or entire applications with just a prompt. Would love feedback, feature requests, suggestions, or comments below! Arnav Agrawal Adityavardhan Agrawal