Skip to content
View kashavpiya's full-sized avatar
🔬
Researching LLM reasoning & fine-tuning
🔬
Researching LLM reasoning & fine-tuning

Block or report kashavpiya

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
kashavpiya/README.md

Hi, I'm Kashav Piya

Typing SVG

AI that serves people first.

Website LinkedIn Hugging Face Email


About me

I'm a Computer Science graduate (Augustana College, '23) now pursuing a Master's in Artificial Intelligence at Indiana Wesleyan University. By day I'm an AI Associate, building automation and AI systems that augment how people work, not replace them.

I've published research as first author in IEEE and arXiv, including work on mitigating bias in voice AI and responsible IoT design in healthcare. I bring that grounding to what I build: RAG pipelines, AI agents and workflow automations that are explainable and purposeful.

  • 🔬 Research: LLM evaluation, reasoning, fine-tuning, AI equity
  • ⚙️ Building: AI agents, RAG systems, n8n workflow automation
  • 🎓 Studying: M.S. Artificial Intelligence, Indiana Wesleyan University
  • 🤝 Open to: research collaborations and interesting automation problems

Featured work

Project What it is
🧠 kas-4b Qwen3-4B fine-tuned with high-rank LoRA plus a custom logit-scoring engine for the Decision Index benchmark. #1 on CRUXEval+, #2 on Habermas Machine+ and BPoMP+, #3 on CLadder+ out of 112 entries, beating models up to 28B. Model · GGUF
📊 clarity-reasoning-effort Does "overthinking" hurt LLMs on subjective tasks? Studies o3-mini, GPT-5.4 and Claude Sonnet 4.6 across reasoning-effort levels on the CLARITY political interview dataset, split by annotator agreement.
💬 Ask-Kashav agent An AI agent trained on my professional background, built with n8n, Claude Haiku 4.5 and LangChain, streaming responses to a custom React frontend.
🏛️ City of Goose Lake, IA Full municipal website, live and used by the community.
🌸 StarFlower Clinic Clinic website: design, development and deployment.

Publications

  • Overthinking Hurts Most When Humans Disagree: Reasoning Effort and Annotator Agreement in Subjective LLM Classification · SSRN, 2026 · paper
  • From Deep Learning to Real-World Surveillance: A Case Study of Flock Safety's ALPR Network · SSRN, 2026 · paper
  • Addressing the Selection Bias in Voice Assistance: Training Voice Assistance Model in Python with Equal Data Selection · arXiv, 2023 · paper
  • Are the Internet Connections at Augustana College Good Enough for Student Productivity? · ICDSST, 2023 · paper
  • Virtual Reality: The Answer to the Future of Gaming · ARSSS Conference, 2022 · paper
  • IoT in Health Care Industry: A Promising Prospect · IEEE, 2021 · paper

Stack

AI & automation

Claude Hugging Face PyTorch LangChain n8n Supabase scikit-learn Pandas

Languages & tools

Skills

GitHub activity

GitHub stats Top languages GitHub streak

Have a project in mind or a research collaboration? Reach out.

Pinned Loading

  1. VoiceAssistant VoiceAssistant Public

    Python voice assistant with speech recognition, from the research on reducing gender bias in voice AI

    Python 2 2

  2. f1_analysis f1_analysis Public

    Formula 1 driver comparison tool: compare lap and sector times using the OpenF1 API

    JavaScript

  3. clarity-reasoning-effort clarity-reasoning-effort Public

    Code and data for the paper: Overthinking Hurts Most When Humans Disagree - reasoning effort vs. annotator agreement in subjective LLM classification

    Jupyter Notebook

  4. kas-4b kas-4b Public

    Decision Index v0.2.1 submission — kpiya/kas-4b (Qwen3-4B fine-tune)

    Python

  5. kashavpiya.github.io kashavpiya.github.io Public

    Personal site source: piyakashav.xyz (React, Vite, Tailwind)

    JavaScript