I'm interested in building reliable, scalable, and aligned AI systems, with a research focus spanning two connected areas:

• Multilingual hallucination detection
• Scientific factuality classification
• Large-scale dataset balancing & augmentation
• Cross-lingual model evaluation
• LLM reliability & safety

• Internal alignment embeddings (ESAI)
• Counterfactual harm forecasting
• Graph diffusion & attention gating
• Zero-shot population scaling
• Stability of cooperative policies

I enjoy working at the intersection of LLMs, distributed training, MARL, and alignment-focused architectures.

📫 Email: [email protected]
🤗 HuggingFace: https://huggingface.co/Haxxsh

Lead Implementer | Dataset Curation + Modeling

• Constructed a unified multilingual dataset (172× expansion, 124k samples).
• Fine-tuned XLM-RoBERTa-Large for hallucination detection.
• Achieved 2nd place (Gujarati zero-shot) and competitive rankings across 9 languages.
• Developed data pipeline, augmentation, evaluation, and error analysis tools.
  🔗 Repo: https://github.com/ezylopx5/SHROOM-CAP2025
  🔗 Paper: https://arxiv.org/abs/2511.18301

Architect | MARL + Alignment Research

• Designing differentiable alignment embeddings that regulate emergent agent behaviors.
• Implementing similarity-weighted graph diffusion & counterfactual forecasting.
• Studying zero-shot population scaling (4 → 16 agents) and causal interventions.
• Paper: https://arxiv.org/pdf/2512.18309
  🔗 Repo coming soon

Top 18% (Rank 3617 / 20,698 teams)

• Built end-to-end multi-modal pipeline (TF-IDF + PCA + ResNet embeddings).
• Ensemble of XGBoost / LightGBM / CatBoost with 7-fold CV.
• Implemented inverse-SMAPE–weighted ensemble.
  🔗 Repo: https://github.com/ezylopx5/AmazonMLHackathon

AI + Web Engineering Project

• Built an AI chatbot, species DB, interactive ocean-depth explorer.
• Next.js + Leaflet.js + multiple data APIs.
  🔗 Demo: https://oceanic-project.vercel.app/

• PyTorch, Transformers, XLM-R, HuggingFace
• Reinforcement Learning (PPO, Actor-Critic, vectorized rollouts)
• Dataset curation, augmentation, multilingual modeling

• CUDA • Mixed Precision Training • GPU Optimization
• Docker • Linux • Git
• Building reproducible ML pipelines

• W&B, TensorBoard, VS Code