Kriti Behl

Kriti Behl

Human-Centered AI Researcher | LLM Evaluation | Multimodal Intelligence

Hi, I design evaluation and reasoning systems that make AI safer for humans. My work focuses on intent-preserving repair, multimodal reasoning, and transparent model scoring.

🔥 Featured Work

FairEval-Suite

Human-Aligned Evaluation of Generative Models (beyond accuracy and toxicity). A lightweight scoring framework using:

  • Helpfulness
  • Relevance
  • Clarity
  • Transparent toxicity signals

Live Demo: https://huggingface.co/spaces/kriti0608/FairEval-Suite
PDF: https://zenodo.org/records/17625268
GitHub: https://github.com/kritibehl/FairEval-Suite

VoiceVisionReasoner

Multimodal reasoning across Speech → Vision → Language. Transforms emotional human signals into collaborative problem solving.

  • Speech transcription
  • Visual grounding
  • Joint reasoning
  • Sanity + tone checks

GitHub: https://github.com/kritibehl/VoiceVisionReasoner

JailBreakDefense

Lightweight LLM jailbreak detection and repair framework. Designed to protect intent instead of censoring users.

Zenodo: https://zenodo.org/records/17625268

Research & Publications

  • “FairEval: Human-Aligned Evaluation Framework for Generative Models”
    Zenodo (2025) — https://zenodo.org/records/17625268

  • “I Didn’t Have a Big Research Lab—So I Built My Own AI Safety Tools From Scratch.”
    Medium (2025) — https://medium.com/@kriti0608

  • ResearchGate Preprint (2025)
    https://www.researchgate.net/…

Experience

DevSecOps / Applied AI — Thales Group

  • Built Flask-based REST APIs and dashboards for real-time resource efficiency.
  • Integrated Prometheus, Kubernetes, PostgreSQL logs into CI/CD insights.
  • Built multi-product evaluation metrics for GPU pools.

Impact: reduced compute waste & improved model troubleshooting visibility.

Skills

  • Large Language Models
  • Model Evaluation
  • AI Safety
  • Python, FastAPI, Flask
  • React, JS/TS
  • Kubernetes, Docker, Prometheus
  • Postgres, Mongo
  • ML research workflows

Contact

📨 kriti0608@gmail.com
🎓 University of Florida — CISE (Graduating Dec 2025)
🔗 LinkedIn — https://linkedin.com/in/kriti-behl
🔗 Google Scholar — https://scholar.google.com/citations?user=hUGBL5wAAAAJ

I build AI that understands humans as collaborators — not failed prompts.

Recent Posts