Preetam Chhimpa

M.Tech Data Science, IIT Roorkee (CGPA: 9.29) · Expected graduation: June 2026

Email · GitHub · LinkedIn · CV (PDF)

Research interests

  • Efficient multimodal learning (vision + language), compute- and memory-aware fusion
  • Adversarial evaluation and robustness of vision-language models under distribution shifts

Current research

Student Researcher, SafeTrip Lab, IIT Roorkee (Jun 2025 – Present)

Compute-efficient multimodal fusion for autonomous driving (multi-view camera → T5-aligned scene embedding).

  • Reduced parameters by 52% using anchor-token pruning, Perceiver cross-attention, and gated view pooling
  • Results on multi-view VQA: EM 51%, F1 70%, BLEU-4 47.1, ROUGE-L 69.1, CIDEr 2.99

Selected projects

GeoAnchors — Compute-efficient multimodal fusion
Repo: github.com/preetam1407/GeoAnchors
Adversarial Evaluation of Vision-Language Models
Aug 2025 – Nov 2025
Repo: github.com/preetam1407/Adversarial-Evaluation-of-VLM
  • Training-free adversarial evaluation on IMP-v1-3B and Qwen2.5-VL-7B
  • Answer flip rates: 75.1% (IMP-v1-3B), 38.4% (Qwen2.5-VL-7B)
  • Counting deviation (mean absolute shift): 6.74, 1.24
DiffGraph+ — Heterogeneous Graph Diffusion
Feb 2025 – Apr 2025
Repo: github.com/preetam1407/DiffGraph
  • Extended DiffGraph (WSDM’25) for heterogeneous graphs with automatic view discovery via Graph Transformer
  • DBLP author classification: Micro-F1 78%, Macro-F1 77%, AUC 93%
Flood Forecasting System
Aug 2023 – Nov 2023
Code · Dashboard
  • Pixel-level inundation mapping for the Narmada basin; historical F1: 0.86 (2001–2012)
Cminusminus — Programming Language & Compiler
Jan 2024 – Apr 2024
Repo: github.com/IITGN-CS327-2024/our-own-compiler-com-piler-t6
  • Compiler toolchain using Lark, WASM backend, unit and integration testing
Waste Segregation System
Feb 2022 – May 2022
Code · Prototype
  • ResNet-50 fine-tuning; 94% validation accuracy; OpenCV + Arduino prototype

Open-source

Experience

Scientific Computing Contractor, Mercor (Jul 2025 – Nov 2025)

Built terminal-based scientific benchmarks for evaluating AI agents; auto-graded tasks with deterministic scoring.

Software Engineer Intern, IntentSignal Systems (May 2024 – Sep 2024)

Built a data-as-a-service platform (Next.js) serving 15M+ B2B records; JWT auth, search, CSV export; reduced page load time by 40%.

Cloud Operations Intern, Patible AI (May 2023 – Aug 2023)

Scraping and ingestion pipelines; 2.6M+ records on AWS (EC2/S3); PostgreSQL ingestion optimization.

Skills

Python, C++, C, SQL · PyTorch, TensorFlow, Hugging Face · Git, Docker, AWS, PostgreSQL

Last updated: Jan 2026
CV · Email