he/him 🌈 Research at @percepta Prev: @StanfordHAI @CornellORIE @MSFTResearch, @IBMResearch, @uoftmie
New York, USA
Joined January 2021
- Tweets56
- Following200
- Followers145
- Likes320
Insanely excited about Percepta and our vision for Agentic OR! The world is full of scarcity, OR is an amazing tool to help manage scarce resources - let's make sure everyone can use it and can use it well in the age of AI!
Connor Lawless retweeted
I've been thinking a lot about AI slop lately, and there are a few thoughts that I wanted to document, not because they are finalized and won’t change, but just in case others find them useful. 1/n
Check out Henry's talk! Really love this project and the doors it opens for doing verifiable and trustworthy optimization with LLMs!
Check out Henry's talk on our upcoming @NeurIPSConf 2026 paper, "FLARE: Verifying MILP Reformulations with LLM-Based Theorem Proving."
Paper: arxiv.org/abs/2608.25220
Code + Benchmark: flare.henryrobbins.com
Led by Henry, with @LawlessOpt and @madeleineudell, using @leanprover.
3
108
Connor Lawless retweeted
I'm hiring for our interpretable world modeling effort at Percepta! The core question: how can we make an agent build a simulator, understand a company, and then design a policy, all from unstructured data? Come work with me and our amazing team.
jobs.ashbyhq.com/percepta/90…
Connor Lawless retweeted
Neo in the Matrix, learns Kung-Fu without "training" by getting the skill uploaded to his brain.
Our new architecture, Spotlight, enables LLMs to grow and gain new capabilities without changing their weights. Check out our work!
1/ What would it take for LLMs to be able to grow their own capabilities?
We introduce spotlight, a memory architecture with growing capacity but with linear computational complexity. We show that this suffices to host evolving ecosystems like software by building Python directly into an LLM.
We also share some very early language modeling results where it outperforms attention and linear attention on long context capabilities!
Connor Lawless retweeted
1/ What would it take for LLMs to be able to grow their own capabilities?
We introduce spotlight, a memory architecture with growing capacity but with linear computational complexity. We show that this suffices to host evolving ecosystems like software by building Python directly into an LLM.
We also share some very early language modeling results where it outperforms attention and linear attention on long context capabilities!
Connor Lawless retweeted
We can't let frontier labs choose their own auditors - or their own questions.
Blessed to be working at a company where the CEO @hirshjain drops "agentic OR" 7 times in an all hands meeting
oh hey look an expert at Agentic Operations Research @LawlessOpt working (and teaching) at an Agentic Operations Research company
Super excited about this paper led by the incomparable Henry Robbins!
You no longer need heuristics to check that your LLM correctly reformulated a MILP problem (or passes a benchmark), we can prove it in Lean for any instance!
An LLM can rewrite your optimization problem to improve solve time, pass every test, and still silently change the underlying problem.
Our new paper introduces FLARE, an LLM agent that uses Lean to prove that a given MILP reformulation is valid across all possible problem instances, not just those used for testing.
On 54 reformulation pairs of NP-hard problems, FLARE correctly classified all (best baseline: 83%) and provided a machine-checkable proof for every accepted reformulation.
AI-generated optimization needs proofs; passing tests isn’t enough.
2
122
Connor Lawless retweeted
It's very easy to arm-chair quarterback healthcare transformation from an office in San Francisco. Much harder to actually assemble the avengers in Akron Ohio and try to fix it.
Very proud of the role @percepta plays in this work. Reach out if you want to get on the ground in Akron with our team
In 2024, we made the unprecedented decision to acquire a hospital system. It was an opportunity to create a blueprint for healthcare transformation to make American healthcare proactive, accessible, and affordable.
@SummaHealth and our Health Assurance Transformation Company (HATCo) revealed their technology transformation stack, partnering with some of our portfolio companies @aidocmed, Clarium, @CommureOs, @poweredbyfabric, @hippocraticai, @judihealth, @transcarent, and @versemedical. @percepta is serving as a transformation partner, unifying the tech stack into a seamless Al-native operating system.
We believe the future of healthcare is health assurance with tech founders and health system leaders working together to reimagine healthcare in the US and beyond.
Connor Lawless retweeted
@LawlessOpt and I are excited to present our #AAAI2026 tutorial on “LLMs for Optimization: Modeling, Solving, and Validating with Generative AI.”
When: Tuesday, Jan 20, 2026, 8:30am–12:30pm (Singapore time)
Where: Garnet 216 (Singapore EXPO)
Optimization is central to planning, scheduling, and decision-making, but deploying solvers requires deep expertise. Our tutorial covers how LLMs can support the end-to-end optimization pipeline (model formulation, solver configuration, and model validation) and highlights open research directions.
Tutorial page (agenda + reading list): conlaw.github.io/llm_opt_tut…
(Connor’s intro slides are shown here.)
Thanks to @leo_bix and @madeleineudell for helping put the proposal together.
CC @RealAAAI
Excited to be talking about this project at INFORMS today (ME31 - 4:00 @ Summit 422)! In this paper we build and do the first user study with a system that combines LLMs with Constraint Programming for personalized optimization modelling (accepted in ACM TiiS, an HCI journal!)
Super excited about this project from my time at @MSFTResearch! We explore how to combine LLMs and Constraint Programming to allow non-expert users to design optimization models in the context of meeting scheduling.
5
652
Have you developed and/or implemented an optimization model to solve a real-world use case? We want to hear from you! tinyurl.com/3ejdr7su
Connor Lawless retweeted
Connor Lawless (@LawlessOpt) talks at #cpaior2024 about a new notion of fairness in clustering to ensure minimum majority representation of groups across clusters, which helps shaping more representative electoral districts, & which may shown to be impossible by an exact approach
Connor Lawless retweeted
In my presentation at the AI / OR workshop, I talked about incentives for collaboration across disciplines, touching a bit on journal and conference publishing cultures.
From later conversations, this seems to be less of an issue in IE departments than it is in business schools.
Connor Lawless retweeted
Continuing the second session of the AI / OR workshop, @LawlessOpt talks about using large language models to replace the #orms expert when formulating optimization models, such as for building a constraint programming model for scheduling meetings
Connor Lawless retweeted
Sharing some papers I enjoyed at the OPT workshop at #NeurIPS2023! sorry I couldn't interact with everyone (had a tight flight schedule).
@BaharloueiSina 's paper on f-ERM.
@LawlessOpt's paper on Fair Clustering.
Todo: Another post on DP, fairness, game theory, bandits papers.
Super excited about this project from my time at @MSFTResearch! We explore how to combine LLMs and Constraint Programming to allow non-expert users to design optimization models in the context of meeting scheduling.
"I Want It That Way": Enabling Interactive Decision Support Using Large Language Models and Constraint Programming
paper page: huggingface.co/papers/2312.0…
A critical factor in the success of decision support systems is the accurate modeling of user preferences. Psychology research has demonstrated that users often develop their preferences during the elicitation process, highlighting the pivotal role of system-user interaction in developing personalized systems. This paper introduces a novel approach, combining Large Language Models (LLMs) with Constraint Programming to facilitate interactive decision support. We study this hybrid framework through the lens of meeting scheduling, a time-consuming daily activity faced by a multitude of information workers. We conduct three studies to evaluate the novel framework, including a diary study (n=64) to characterize contextual scheduling preferences, a quantitative evaluation of the system's performance, and a user study (n=10) with a prototype system. Our work highlights the potential for a hybrid LLM and optimization approach for iterative preference elicitation and design considerations for building systems that support human-system collaborative decision-making processes.
