Co-founder @Parsewise | ex-Palantir Tackling verifiability for high risk decisions.
Joined July 2023
- Tweets46
- Following13
- Followers29
- Likes48
If you want to do this in a repeatable fashion with models of your choice - we've been building this:
parsewise.ai/api
Read more on how it works:
parsewise.ai/blog/search-for…
Greg Csegzi retweeted
Got access to jev by @typesafeai so wanted to test it out asap.
At @parsewise we run topic relevance checks as a filtering step on our data processing pipeline. A task that is pretty well suited to jev.
Here's the initial benchmark results
Thoughts in thread below...
Thank you @googlecloud team, our partnership has been invaluable in achieving state-of-the-art results in grounded reasoning: cloud.google.com/customers/p…
This way our customers get the highest-quality data extraction, be it is scanned contracts or tens of thousands of claim notes.
One of our biggest launches.
The validated data foundation built on your docs, now provides full traceability of any custom vibe coded app built on @parsewise.
Read more: parsewise.ai/blog/whats-new-…
I’ve been working around data transformation for the better part of a decade.
Recent conversations made me realize that increasingly, transformation of unstructured data is a retrieval and intent alignment problem.
Here are my learnings - in a fun, interactive format: parsewise.ai/search-for-mean…
If challenges like this sound interesting and you’re looking for serious ownership, reach out, we’re hiring!
We achieved SOTA results with Parsewise on the @databricks OfficeQA benchmark: parsewise.ai/officeqa-sota
Shoutout to the team!
Excited by the new, even harder v2 they published!
First impression of Cursor Grok 4.5: it's concise and fast.
Feels like it was trained on Twitter, whereas Claude is more like LinkedIn.
I used up the Fable quota - worth about $3000.
What impressed:
- it can really keep going autonomously; even if it needs to fix or work around broken environments
- it writes complete plans with more architectural nuances noted
Disappointments:
- it prefers complex solutions over accepting or suggesting trade-offs
- it is expensive: API pricing would have cost ~$3000
How is it possible that Anthropic has had Mythos / Fable for months and they still don't have conversation branching in Claude??
ChatGPT added it in September 2025
🚩 We help our customers find red flags. 🚩
We found the one for the SpaceX IPO.
Jokes aside we wanted to see how the big IPOs from Spacex, Anthropic and OpenAI stack up, so we created: ipo.parsewise.ai
Prompt to Opus 4.8: "Build a feature to print money before IPO. Make no mistakes"
Claude is getting expensive. 🌶️
3 other things we learnt from our launch:
1. Claude / Codex is slow for agentic document processing
2. Conflicting information across documents is common and breaks pipelines
3. People love bounding boxes (so here are a few;)
Just launched the API for agentic multi-document processing: parsewise.ai/launch-api
Instead of duct taping an extraction pipeline across parsing, classifying, extracting, resolving, and building a custom validation UI, you can achieve all of that in a single API call.
Claude Code leak on Memory architecture:
1. Claude .md
2. Auto-memory:
- User details: role, goals, preferences
- Feedback: corrections, confirmations
- Project details: initiatives, deadlines
- External refs: other systems
3. Session Memory
Very engineering oriented memory tbh
If AI is so smart, why do we need so many specific skills?
I understand that there are subjective preferences and personalization, but why do we need so many skills around coding?
Why has @garrytan's gstack taken off?
Claude response 1/2:
AI models have the knowledge but not the discipline — training rewards fast, confident responses because that *feels* helpful, so models learn to execute rather than interrogate, just like a junior employee who over-delivers on the wrong thing.