AI Reasoning powered by Mathematical Logic
Austin, TX
Joined April 2025
- Tweets47
- Following12
- Followers54
- Likes10
We're launching SpecLogician today!
A code diff tells you what changed in the text - not what changed in the behavior.
SpecLogician gives coding agents & humans a shared, machine-checkable model of software behavior, so edge cases surface before production.
Visit speclogician.dev to get started!
More details here: medium.com/imandra/coding-ag…
Join us this Thursday at 10 am CT!
We'll model a slice of exchange behavior end to end: an LLM driving SpecLogician to propose states, transitions, and scenarios, with ImandraX reasoning over the result and surfacing the counterexamples neither of us thought of.
Register here: riverside.com/webinar/regist…
A brokerage that disintegrated in 45 minutes. An exchange that botched its own IPO. A flight control system with behavior its pilots didn't know about.
Different industries, same root cause: an edge case nobody saw coming.
In this new demo, our co-founder @DenisIgnatovich shows how to give @OpenAI's Codex access to ImandraX through CodeLogician, so it can reason rigorously about every corner of its own code.
youtu.be/ZMzcSSYHnkQ
LLMs generate code well. They can't exhaustively reason about a program's state space and missed edge cases are what bring down trading systems, trigger compliance failures, or worse.
Imandra closes that gap!
Watch @GeminiApp :
-Generate a non-trivial Python program
-Hand it to ImandraX for region decomposition
-Discover edge cases it didn't anticipate
-Rewrite the original to handle them
The same approach scales from a single matching function to the protocol specs hedge funds and brokers certify against when connecting to an exchange.
This is what rigorous AI-generated code looks like.
Watch the full video here: youtu.be/djQzK8Ymuao
Imandra's Co-Founder Dr. @PassmoreGrant is keynoting the PLSEFM + AI Workshop at Georgia Tech on April 30, hosted by @GTRI_Global, @Google, and @ORNL.
He'll show how Region Decomposition, a core technique behind Imandra's automated reasoning platform, serves as a unifying data structure for neurosymbolic reasoning. When paired with LLM-based coding assistants, it meaningfully beats LLM-only baselines on edge case discovery, test generation, and formal verification.
Grant also joins the closing panel with @Google's Dr. Jose Cambronero and @GTRI's Dr. Shelby Allen.
Details & registration here: sites.gatech.edu/plsefm-ai-w…
Don't miss it!
Last week on AI Reasoning Weekly, @DenisIgnatovich walked us through real world applications of CodeLogician and how formal verification catches what LLMs miss.
Plus a first look at something coming very soon!
Watch the full video here: youtu.be/DOOcJYitKTg
We ran Claude + CodeLogician on a real @stripe payment flow.
@claudeai drove the loop. ImandraX was the auditor.
Real bug found. Flawed verification goals caught. 10 tests became 84.
Read the full case study by Imandra's Co-Founder & CEO @DenisIgnatovich here: medium.com/imandra/managing-…
Add one line to a bank account classification model and your logical regions jump from 19 to 55. That is a 190% increase in behaviors.
@DenisIgnatovich and Paul Brennan walk through a real-world banking scenario showing how AI-assisted coding via @cursor_ai, augmented with Imandra's automated reasoning, delivers provably correct results.
Watch the full video: youtu.be/hL_Q4Q4GCJQ?si=jsLn…
Free to try: codelogician.dev
We just wrapped our first AI Reasoning Weekly and it was a strong kickoff!!
Here's what we covered 🚪
If you’re building with LLMs and care about correctness, formal reasoning is the missing layer.
Watch the full session: youtu.be/nA9mEGH1lEk?si=60LL…
Next session: Thursday March 12, 10am CT. Registration link below.
