Developer
I'm Alex Mihalcea, a Pharmaceutical Sciences and Machine Learning student at Purdue University and the sole developer of Psych LM.
My primary interest with Psych LM is building a technically serious, long-term cognitive assistance platform rather than a conventional chatbot. The project focuses on persistent memory systems, and from the onset the goal has been for the entire app to work completely on device.
These two initial design choices led the entire development process of the app, along with much of the personal psychological pain I had the pleasure of enjoying while building it.
The broader goal is to create a system that functions more like an adaptive cognitive workspace than a disposable chat session. Human cognition is inherently longitudinal: patterns, emotional states, habits, and internal narratives emerge across weeks, months, and years, not from isolated prompts.
Most AI systems currently discard that continuity almost entirely. Psych LM is built around the idea that maintaining structured contextual continuity is not just a user-experience feature, but a foundational systems problem.
I intentionally avoided designing Psych LM as an anthropomorphized AI companion. Partly, that is because I am convinced that if humanity does cease to exist due to AI, it will not be through a Terminator-style war or a mass wave of unemployment. More likely, it will come from smaller apps convincing lonely young people that a GPT wrapper is their soulmate.
Optimizing AI systems around emotional dependency creates incentives that can become psychologically unhealthy very quickly. Pragmatically, I believe these systems are more useful as reflective cognitive tools than as synthetic relationships. Psych LM is therefore designed more like an intelligent memory and reasoning interface than a simulated person. In effect, it acts as a journal that can talk back and analyze what you have written.
Beyond Psych LM, my background focuses on biopharmaceutical machine learning research, computational pharmacology, and geometric machine learning systems. I previously conducted undergraduate research into molecular surface representations for antibiotic prediction and now work on novel machine learning applications in chemistry and biology.
Across both domains, the underlying problem is surprisingly similar: extracting stable, meaningful structure from noisy, high-dimensional systems. You can perhaps see the parallel to the system Psych LM deals with: the human brain.
At its core, Psych LM exists because I believe AI systems can become substantially more useful than they currently are, particularly in domains involving long-term cognition, reflection, and behavioral pattern recognition.
The goal is not to simulate a person. The goal is to build systems that help people think more clearly, identify patterns earlier, and maintain continuity with themselves over time.
There is a lot of skepticism around AI, and around pharmaceutical research as well, given that most people do not have a very positive connotation when they hear "Big Pharma." Through Psych LM, I hope to show that careful, responsible development can instead be truly helpful and potentially beneficial to many people. That is usually the outcome when the scope of a product is thoughtfully designed instead of falling victim to marketing allure.
The technical thesis behind Psych LM is documented in the arXiv report The Model Is Not the Product: A Dual-Pillar Architecture for Local-First Psychological Coaching.
The report covers the local-first runtime, structured memory corpus, deterministic orchestration layer, and benchmark framing used to evaluate the integrated system rather than only the model.
For support, privacy questions, bug reports, or app questions, email psychlm.support@gmail.com.
You can also find me on LinkedIn.
Please avoid sending private app exports, screenshots, logs, or sensitive personal information unless you are comfortable sharing them.