Building @GraphonAI. Alum @umbc @msbgu. community @hackersnbackers
San Francisco, CA
Joined January 2022
- Tweets135
- Following332
- Followers300
- Likes417
Over a year ago, I left VC to build Graphon with @graphonfounder and @clarkzhango. Today we’re announcing our $35M Series A, six months after our seed.
Models keep getting smarter, but they still forget. Context and persistent intelligence is the core missing block in the AI infrastructure puzzle.
The benchmarks speak for themselves: #1 on ATM-Bench, 94% on LoCoMo and a new cost-accuracy frontier across 7 benchmarks. What I’m proudest of is what happens outside the benchmarks: enterprise customers running this in production, at 90%+ precision and a fraction of the cost.
Thank you to the customers, investors and team who believed early.
We are just getting started.
Try it: app.beta.graphon.ai
After years working on robot memory at @NASAJPL, @GoogleDeepMind , @Meta and @nvidia, @clarkzhango and I kept asking: why does AI still forget?
Today, with @deepakmishraVC, we’re announcing @GraphonAI and our $35M Series A just 6 months after our seed.
• #1 on ATM-Bench, ATM-Bench Hard
• 94% on LoCoMo
• 84% on DolphinBench, 10x cheaper
• A new cost-accuracy frontier across 7 benchmarks
businessinsider.com/graphon-…
SF Tech Week is on and Graphon got you covered ☕🚗
Free coffee pop-ups:
Tue, Dogpatch: partiful.com/e/YmLLByo9kUXgD…
Thu, 51 Federal Street: tech-week.com/calendar/sf/ev…
Free rides to evening events, Mon-Fri: luma.com/xll5ryfk
Come say hi 👋 #SFTechWeek
@techweek @GraphonAI
Literally unlocking net new capabilities in multimodal memory with @GraphonAI @typesafeai
I mapped an entire movie by mood for about 5 cents (Jev @typesafeai + @GraphonAI).
Moods like: Funny, Scary, Thrilling, Suspenseful etc.
Around 7,000 AI analyses for roughly $0.05 and 7 seconds.
The demo is pretty simple:
First, Graphon maps the movie.
We ask Graphon about 40 questions designed to cover what's happening throughout the video. Graphon returns descriptions of the scenes, along with citations to the exact moments in the movie.
We use those questions to build a picture of what's happening across the entire movie.
Then we send those scene descriptions to Jev from @TypeSafeai
Jev looks at each scene and judges the probability that it is funny, scary, suspenseful, exciting, calm, or whatever other categories we want to analyze.
That's where we do around 7,000 analyses in parallel.
The result is basically a mood map of the entire movie.
Now imagine you're a movie studio, video editor, or creator.
You could say:
“Show me every scary moment.”
“Find all the car chases.”
“Give me the funniest 30 seconds.”
“Create a compilation of the most suspenseful moments.”
And jump directly to those exact scenes.
Today, doing this at scale is surprisingly complicated. Someone has to watch, tag, classify, scrub through footage, and manually find all of these moments.
With Graphon + Jev, you can start doing it automatically.
And the crazy part?
Those ~7,000 mood analyses cost about 5 cents.
This is what I think multimodal search starts to look like when you combine understanding, citations, and reasoning across an entire video.
@CompleteSkeptic what do you think?
Deepak Mishra retweeted
I mapped an entire movie by mood for about 5 cents (Jev @typesafeai + @GraphonAI).
Moods like: Funny, Scary, Thrilling, Suspenseful etc.
Around 7,000 AI analyses for roughly $0.05 and 7 seconds.
The demo is pretty simple:
First, Graphon maps the movie.
We ask Graphon about 40 questions designed to cover what's happening throughout the video. Graphon returns descriptions of the scenes, along with citations to the exact moments in the movie.
We use those questions to build a picture of what's happening across the entire movie.
Then we send those scene descriptions to Jev from @TypeSafeai
Jev looks at each scene and judges the probability that it is funny, scary, suspenseful, exciting, calm, or whatever other categories we want to analyze.
That's where we do around 7,000 analyses in parallel.
The result is basically a mood map of the entire movie.
Now imagine you're a movie studio, video editor, or creator.
You could say:
“Show me every scary moment.”
“Find all the car chases.”
“Give me the funniest 30 seconds.”
“Create a compilation of the most suspenseful moments.”
And jump directly to those exact scenes.
Today, doing this at scale is surprisingly complicated. Someone has to watch, tag, classify, scrub through footage, and manually find all of these moments.
With Graphon + Jev, you can start doing it automatically.
And the crazy part?
Those ~7,000 mood analyses cost about 5 cents.
This is what I think multimodal search starts to look like when you combine understanding, citations, and reasoning across an entire video.
@CompleteSkeptic what do you think?
Deepak Mishra retweeted
At Graphon, we believe intelligence lives in the connections, not the nodes.
So we are bringing the graph to the table.
We are hosting our first Engineering Leaders Dinner sponsored by J.P. Morgan with confirmed attendees from OpenAI, Anthropic, Harvey, Carta, Google, Hippocratic, Lucid Motors, Prime Intellect, Mercor and more.
We have a few seats left and if you are passionate about topics like context intelligence, memory, knowledge graphs, multimodal data and LLMs would love to have you join 👇
luma.com/k7nnf5g6
#MultimodalAI #EngineeringLeaders #Agents #ContextIntelligence #Memory #Graphon
Five years in the making.
I spent five years evaluating data and AI companies before deciding to build one. The same ceiling kept showing up - capable models, solid retrieval, but unreliable results. Not because the models were bad. Because they destroy structure before reasoning even begins.
Most systems chunk data into tokens, embed them into vectors, and retrieve by similarity. By the time the model sees anything, the context is already broken.
So we built Graphon, a graph-native intelligence layer. Instead of flattening your data, we convert it into a graph that preserves structure across text, video, audio, and images. Then we compress it down to exactly what is relevant to your query so the model only ever sees a tiny, high-signal slice rather than a noisy pile of chunks.
The result is much better retrieval, cheaper reasoning, and far more accurate results. We are already deployed with some of the largest enterprises and AI-native companies in the world, and we are just getting started.
Read more in @WSJ [wsj.com/tech/ai/graphon-says…]
Try Now: graphon.ai/waitlist
Graphon is out of stealth today. $8.3M in seed funding.
We’re building the pre-model intelligence layer — the layer that sits before your AI model and maps how data connects.
Every model has a ceiling on what it can process. Your data shouldn’t.
More in @WSJ: wsj.com/tech/ai/graphon-says…
Learn more: graphon.ai
Billions in value created.
Ran into @sama at dinner…
A week later: @openclaw acquisition headlines
Silicon Valley never disappoints @steipete
Wife @sachiangle and I interrupted @sama at dinner and he was so nice about it! Would love to show you how @GraphonAI has solved for long context multimodal ai Sam! 😄
9
571
Deepak Mishra retweeted
We talked to dozens of Fortune 500 AI leaders while building Graphon.
One thing has become clear: enterprise AI doesn’t need bigger models, it needs better architecture.
Systems built for text break under multimodal complexity. Teams aren’t leaving for marginal gains; they’re leaving because current architectures can’t keep up.
A quiet inflection point is here.
CC: @ArbaazK87777726, @deepakmishraVC
Wife @sachiangle and I interrupted @sama at dinner and he was so nice about it! Would love to show you how @GraphonAI has solved for long context multimodal ai Sam! 😄
We were pleasantly surprised by the engagement from the developer community. Thanks @agihouse_org @GoogleDeepMind for being great co-hosts.
Stay tuned for what’s coming next year 👀
Hello world, we’re @GraphonAI
We showed up to @agihouse_org’s Gemini Build Day to see if devs would care…
Turns out 1 in 4 teams built on Graphon 👀
• “Better than other model providers on factual queries and quality” (their words, not ours)
• Used alongside Gemini for things like drug discovery
• $1k prize → tie between two Graphon-powered teams
All projects here: app.agihouse.org/events/gemi…
More to come from us next year 🙂
3
250
Come build on top of @GraphonAI's Relationship Systems Models. Pumped to join @agihouse_org @Google @yutori_ai at Gemini 3 Build Day on Dec 13.
Join our Gemini Build Day on 12/13 at AGI House with @Google, @GraphonAI @yutori_ai, and @agihouse_org.
🤖 Build with Gemini 3, Google’s newest multimodal model
🛠️ Create agents, multimodal copilots, and AI-native workflows
🚀 10 hours of hacking, demos, and rapid prototyping
🏆 50k / 35k / 15k Google Cloud Credit prizes
👥 Builders, researchers, founders, and engineers
RSVP: app.agihouse.org/events/gemi…
Deepak Mishra retweeted
🚨New Paper Alert! 🚨
Our ICML 2025 paper (led by @Narmeen29013644) shows how small models can help steer and align much larger ones by building “bridges” between them.🧵👇
Way to go @withmartian🚀🚀
Excited to make two announcements today! --
1. We're partnering with Accenture to power their LLM Switchboard and >$1B in Gen AI Deployments.
2. We're launching Airlock, our LLM Compliance Automation tool.
2
416
We would love to connect anyone building in the space
14/
@Hitachi ventures shared their market map covering AI agents.
Great to see all our peers @getlindy @MultiOn_AI @griptapeai @sema4ai @browserbase @ReworkdAI @wordware_ai @langchain @pinecone @AdeptAILabs + others
xitter.gq/chiefaioffice/status/1…
1
282
Thanks
@chiefaioffice. Check out the detailed post here: linkedin.com/posts/dmishra91…
4
375
Deepak Mishra retweeted
Hackers & Backers hosted another remarkable event, this time on Data Center Energy Management. Here are a few key takeaways that stood out:
Deepak Mishra retweeted
OpenAI's top safety researchers quit last week. It's easy to say this is OpenAI's fault, but it's not -- it's incentives.
Capitalism and safety aren’t aligned. And capitalism always wins.
To make progress on safety, we need to align those incentives. Here’s how we do it 🧵
Deepak Mishra retweeted
Neural networks are notoriously opaque. Let’s change that.
At Martian, we’re developing methods to make black-box models transparent, editable, and verifiable. Called “model mapping”, it powers 1K+ businesses through our model router.
A sneak peek of what we’re working on:
Let’s goooo
🎉 Martian was named one of the top 100 AI companies in the world by @CBinsights! 🎉
It’s an honor to join the likes of @OpenAI, @databricks, @perplexity_ai, and others in 2024’s AI 100.
1
176