Diptanu Choudhury
@diptanu
Followers
3K
Following
3K
Media
152
Statuses
8K
Founder @tensorlake. Past - AI and Distributed Systems at @meta, @hashicorp, @linkedin and @netflix
San Francisco, CA
Joined December 2007
Excited to announce @tensorlake Cloud! đ§”. Tensorlake converts real-world documents into clean, structured data for business workflow automation and for building Agents in mission-critical documents. It's powered by a state-of-the-art document layout understanding model trained.
Announcing Tensorlake Cloud. Up-leveling Document Ingestion and Workflows for building agentic applications and complex business workflows.
15
19
71
I find it interesting Polars calls their new cloud âserverlessâ but they launch the compute infrastructure insides customers clouds. I am guessing their pilot customers asked them for data residency. This is a challenge at the moment for AI/ML data companies.
1
0
1
Event Driven systems are great for scalability. I have seen them get hard to maintain over time as more events are added to add new features. Have to reason about how the application state changes as various permutations of sequence of events are applied. I like LangGraphâs.
This is langgraph. Most people think of langgraph as agent abstractions, but itâs powered by a low level event driven framework under the hood. If we exposed that - would people be interested? Or focus more on agent abstractions?.
0
2
11
The algorithm really awards shit posting on emerging topics. Is there an AI agent to automate slop production?.
0
0
2
We run 4 different models for parsing documents in @tensorlake's document ingestion pipeline. We implemented dynamic scale outs in our workflow engine to make the bigger and slower model eat up GPU capacity in our cluster to run ingestion faster!
0
2
4
Terraform wouldnât have replaced Puppet/Chef if cloud-native primitives like ELBs and subnets didnât become essential. I donât see Terraform being replaced for what it does well. But the future lies in abstractions that let devs ship workflows(and services) without thinking about.
Terraform is still the best. But I'd like to see someone replace it. The major alternatives aren't interesting to me cause they're too iterative and copycat. I want to see fundamentally new ideas take hold. IaC feels stagnant.
0
0
2
This is the first 100% AI generated PR in @tensorlake open source compute engine - . Explained the model that we need to kill functions in containers with no pending tasks, and if killing enough functions is not going to help with fitting new functions,.
0
0
5
We are solving the checkbox detection problem in documents for good. We will ensure consistent detection of inline checkboxes and the ones which appear inside tables and forms. If you are parsing LONG documents with 100s of checkboxes please ping me, would love to partner.
0
0
1
The infrastructure to keep indexes fresh is not easy to build or operate -.1. Got to stage ingested raw data on blob stores before they are processed. SSDs are not cheap. 2. Heterogeneous compute infrastructure of GPUs and CPUs since ETL for search is now dependent on LLMs and.
If you are using search tools offered by closed LLM providers (e.g., Anthropic, OpenAI), be aware that these companies use their own search indexes, which are not updated in real-time or even regularly. As a result, you may occasionally encounter 404 links, which can tarnish the.
0
0
3
Come hang out with us at AWS Loft if you are in San Francisco! . I will do a talk that will show the journey of a stateless Agent to becoming stateful with Memory, with Tensorlake, Qdrant and n8n.
0
0
2
This tbh. Back in the day product launches of dev tools used to be benchmarks people would argue for a day, now its a video which could pass as a Super Bowl ad đ.
0
0
2
I wanted to post a video of the app, but Apple is not letting me add the Authentication with Apple capability to this until I register as a developer (which takes 48 hours to process) :( . I want to push this to the App Store, and keep working on this to see how far this tech can.
0
0
0
My impression of Claude Code - .1. Building apps with this is totally viable. I would go as far to say this performs at the level of a mid level software engineer. 2. It doesn't come up with the best algorithms. But it's good at following instructions if you spell it out. 3.
Some hiccups - it generated corrupted Xcode project files. I expected it to use command line tools for something like this and not hallucinate or generate these files using a language model.
3
0
9
Some hiccups - it generated corrupted Xcode project files. I expected it to use command line tools for something like this and not hallucinate or generate these files using a language model.
Immediately feeling the need for âvibe designâ which gets me Figma designs of the app, and that obviously needs integration with Claude Code đ
.
1
0
0
Immediately feeling the need for âvibe designâ which gets me Figma designs of the app, and that obviously needs integration with Claude Code đ
.
Day 1 of vibe coding. Building myself an EA, who can take notes and remember stuff about me. Simple stuff. No integration with tools or anything.
0
0
0
Day 1 of vibe coding. Building myself an EA, who can take notes and remember stuff about me. Simple stuff. No integration with tools or anything.
1
0
6
Foundations of Computer Vision - . Nice to see them explain CV in an approachable way!.
0
0
0
RT @LangChainAI: đđ€ Tensorlake Document Processing. Tensorlake integrates with LangChain, enabling LangGraph agents to transform unstructurâŠ.
0
32
0
Itâs incredible to see the adoption of AI developer tools among Product Managers. The AI platform shift is different in this aspect than J2EE/EJB, Web2.0, Mobile, Containers and Cloud. Business stake holders being involved early, is speeding enterprise adoption.
0
0
2
RT @LangChainAI: đ€LangGraph+ Tensorlake: Unlocking Document Understanding for Agents. When creating agents that interact with data, the conâŠ.
0
42
0