Compact AI.
Real Impact.
Open Research.
HeuristixAI is an independent AI research initiative publishing open work on efficient language models, applied intelligence, and human-centered AI systems.
Latest Research
Open work, shared freely.
Every paper we publish is available publicly on ResearchGate and HuggingFace. No paywalls.
Dual-Path Disagreement Resolution: Teaching Small Language Models to Argue and Self-Adjudicate via QLoRA Fine-Tuning
We introduce a dual-path disagreement resolution schema: training compact models to generate competing answers, identify the logical conflict between them, and self-adjudicate to a final justified conclusion.
Low-Resource Self-Reflective Fine-Tuning of Compact Language Models Using LoRA
Investigates how self-reflection capabilities can be instilled into compact language models via structured LoRA fine-tuning, demonstrating meaningful self-correction with significantly reduced compute.
Enhancing the Efficacy of Generative AI through User-Centric Prompt Optimization
Proposes a user-centric framework for prompt optimization that bridges the gap between human intent and generative AI output, democratizing access across all skill levels.
What We Do
Three things we actually do well.
Efficient AI Research
We fine-tune compact models to punch above their weight, publishing open research on LoRA, self-reflection, and low-resource ML techniques.
Open Publication
All research is published openly on ResearchGate and HuggingFace. No paywalls. Built for the community.
Applied AI Builds
From GeniePrompt to AI security systems: we don't just theorize. We build things that work and ship them.
HAI-DualPath-0.5B
A tiny model that argues with itself to reach better answers
Our latest open-source model, trained via QLoRA to generate two competing answers, identify the conflict between them, and resolve it into a justified conclusion. Structured dual-path schema. Verified through ablation. 100% format adherence on evaluation prompts.
HAI-DualPath
Dual Reasoning · QLoRA
Open Source · Free
HAI-ReflectMini-0.5B
A tiny LLM that thinks twice
Our first open-source model, fine-tuned with LoRA to exhibit self-reflection. Generates an initial answer, critiques it, then produces a revised, improved response. Fine-tuned on a structured reflection schema with verified self-correction behaviour.
GeniePrompt
Optimize any AI prompt in seconds
Our first live AI product. Paste any rough prompt and get back a structured, optimized version instantly, grounded in our published research on user-centric prompt optimization.
Before
"write about AI"
After ✦
"Write a 900-word explainer for a tech-savvy audience..."
From the Co-Founder's Desk
Junaid Khan
Co-Founder & Head of Operations
"We started HeuristixAI with a simple belief: that open, efficient AI research shouldn't be locked behind institutions or paywalls. Every paper we publish, every model we release, is our way of giving back to the community that made this work possible."
Internship Program
We work with curious minds.
HeuristixAI runs an active internship program for students passionate about AI research and development. Real work. Real credit. Real certificates.
View Open Roles