Saif Punjwani
Hey! I'm Saif 👋
Incoming M.S./Ph.D. @ Columbia • LLM reasoning • Multimodal agents • Systems
🌻 Currently working on LLM reasoning (Weight‑of‑Thought), prepping for Columbia, and tinkering with evaluation tools.
Research interests
Incoming M.S./Ph.D. in Computer Science at Columbia University. Previously, Joint BS/MS in CS at Georgia Tech (4.0/4.0). I work on reasoning in language models, multimodal agents, and practical systems that ship.
I like simple designs, reliable software, and crisp interfaces. If you value thoughtful, pragmatic engineering, we’ll get along.
Experience
-
Research Scientist Intern — General Robotics
Built real‑time perception→grasp→motion on UR5e; fused RGB‑D and segmentation; trained hierarchical RL (BC→SAC/HER); added program‑of‑thought planner with verifier for robust sim→real.
-
Research Intern — Scale AI
Designed a multimodal reasoning‑agent evaluation harness; built planner→actor→checker agents; added adaptive compute routing and batched inference to improve success and reduce cost.
-
Lead Research Engineer — Barometer
Led Host Intelligence engine (PyTorch/C++), real‑time news analysis; built video analysis (TensorFlow/OpenCV/AWS) used by 20+ media companies; accelerated pipelines with Kafka/Spark.
-
Quantitative Research Intern — Trading Firm
Developed mid‑price prediction on limit order books; feature engineering with stochastic processes and deep models, improving accuracy by ~15%; built a strategy visualizer for alpha discovery.
-
Software Engineering Intern — Siemens
Built LSTM classifiers for breaker signals (≈99% accuracy); developed BLE communications backend (C++/PyTorch) with a React UI; helped launch internal version‑control workflows.
Research
-
Undergraduate Researcher — NLP X Lab
Exploring LLM reasoning (Weight‑of‑Thought) inspired by cognitive science; building benchmarks for interpretability, consistency, and multi‑step inference.
-
Undergraduate Researcher — AI Voice Assistant Lab
Pioneered Large Body Language Models for real‑time gesture generation; developed Allo‑AVA dataset; integrated symbolic reasoning with neural architectures.
-
Research Scientist Intern — Stanford SAIL
Researched compositional grammar‑guided attention improving zero‑shot generalization; geometric deep learning foundations; CRFs for kinematic trajectory prediction.
-
Research/Software Engineering Intern — Columbia University Medical Center
Built C++/Python sensor pipelines; designed CNNs for analyzing neural sensor fracture patterns; engineered microcontroller firmware for data collection.
-
Research/Software Engineering Intern — Structural DNA Nanotechnology Lab, NYU
Built TensorFlow models to predict DNA crystal thermodynamics; streamlined ML regression for nanofabrication; co‑authored ACS Nano publication.
Publications & Preprints ✒️
Weight‑of‑Thought Reasoning: Exploring Neural Network Weights for Enhanced LLM Reasoning
* If you have a preferred ordering or more entries, I can update.