Presenting The Perceptual Observatory at Voxel51 Best of WACV (30 Apr). Project page →
Fenil Bardoliya
I study how language becomes structure: text-to-table generation, information extraction, and multimodal reasoning—building systems that produce interpretable tables and graphs from text and measuring how LLMs and MLLMs behave under distribution shift, augmentation, and real deployment constraints.
I work at the Complex Data Reasoning & Analysis Lab (CoRAL) with Dr. Vivek Gupta. My recent projects include schema-guided text-to-table pipelines, interactive NL-to-SQL systems over sports data, and large-scale evaluation of vision–language models for grounding and robustness.
At ASU, I worked in DREAMS Lab under Prof. Jnaneshwar Das, interned at Samsung Semiconductor India Research, and completed my B.E. in Computer Science at BITS Pilani. I care about reproducible benchmarks, clear failure modes, and tools researchers can actually use.
🔬 Research interests
- 📊 Text-to-table
- 🧾 Information extraction
- 🖼️ Multimodal LLMs
- 🗄️ NL2SQL & tools
✨ Open to full-time roles · Available immediately
📰 News
🤝 Service
- Reviewer — ICML, CVPR, ACL workshops; CVPR 2026 (CogVL); ACL 2026 (SURGeLLM); ICML 2026 (AI4Science).
- Teaching & mentoring — CSE 576 NLP (mentor, informal TA), CSE 575 grader, thesis mentor — ASU.
📚 Publications
Loading…
No publications loaded.
💼 Experience
NLP Researcher Complex Data Reasoning & Analysis Lab (CoRAL)
- Built a scalable evaluation framework for MLLMs under 30+ perturbations across 62K+ samples (scalable to 1M+), supporting deployment-facing reliability analysis.
- Designed Map&Make, a schema-guided and agentic text-to-table pipeline; evaluated frontier LLMs with 5+ structural and semantic metrics.
- Implemented distributed post-training with LoRA, DPO, and GRPO on A100/H100/H200 clusters, improving GSM8K and MATH by 20-40%.
- Co-developed SPORTSQL over live EPL data; contributed 1,793 benchmark queries and reached up to 80% exact match and 94% LLM-as-judge accuracy.
- Built annotation and quality-control protocols for human and LLM-assisted evaluation, improving reproducibility and failure analysis.
Computer Vision Research Aide Distributed Robotic Exploration and Mapping Systems (DREAMS) Lab · Arizona State University · Prof. Jnaneshwar Das
- Developed 3D reconstruction and scene-modeling pipelines for geological formations using Gaussian Splatting and NeRF.
- Recreated Apollo lunar landing sites for immersive simulation environments, improving spatial fidelity and neural-rendering robustness.
Assistant Engineer Samsung Semiconductor India Research
- Built automation pipelines for network log capture and analysis.
- Developed an LSTM-based anomaly detector for SIP and IMS traces over VoLTE, reaching ~80% accuracy across 25+ failure modes.
🚀 Projects
Vermillion: Text-to-Video
- Improved VAE bottleneck in an in-house T2V model.
- Studied latent compression vs reconstruction/temporal quality trade-offs.
Exploring Unlearning in SSMs
- Compared targeted unlearning in SSMs vs Transformers.
- Benchmarked with Perplexity, BLEU, ROUGE-L, and BLEURT.
Dark Motions
- Proposed a framework for extreme low-light video restoration.
- Targeted noise suppression, detail retention, and frame stability.
Medical Imaging Pipeline
- Built X-ray classification, segmentation, localization, and clustering workflows.
- Reached strong metrics on MiniJSRT and extended to larger baselines.
🛠️ Tech stack
Languages
Python
C++
-
Java
JavaScript
TypeScript
Kotlin
Bash
R
SQL
C
HTML5
-
CSS3
XML
ML, frameworks, and data
PyTorch
TensorFlow
Hugging Face
Scikit-learn
OpenCV
Jupyter
Pandas
NumPy
GraphQL
FastAPI
Flask
React
Node.js
LangChain
MLflow
Keras
spaCy
SciPy
Plotly
Postman
Databricks
Databases
PostgreSQL
MongoDB
Redis
Elasticsearch
MySQL
MariaDB
Firebase
-
SQL Server
Tools & infra
Docker
Git
GitHub
GitLab
-
AWS
GCP
-
Azure
Kubernetes
Terraform
-
VS Code
Linux
LaTeX
GitHub Actions
W&B
Wireshark
JUnit
Ansible
REST APIs