Skip to content

Résumé

Aakash Madabhushi

Available from December 2026, in the San Francisco Bay Area, and open to relocating.

aakash.vardhan15@gmail.comlinkedin.com/in/aakash-vardhangithub.com/aakashvardhan

Last updated September 2026

Download PDFSee my workLinkedIn

About

I'm a Master's student in Applied Data Science at San José State, graduating December 2026, and a Graduate Research Assistant in the Department of Applied Data Science, where I train and evaluate manipulation policies on a physical robot arm. I also work as an Instructional Student Assistant for machine learning.

Before grad school I spent a year as a software engineer at Sapaad, building multi-tenant data pipelines that supported analytics for more than 75 customer dashboards at over five million rows a day. Before that I was a computer vision intern at Toothlens, building an unpaired image-to-image pipeline that generated orthodontic previews without the paired patient photos the clinic had no way to collect. My undergraduate degree is a B.S. in Artificial Intelligence from Illinois Tech.

The thread through all of it is that I like problems where you have to build the measurement apparatus before you can tell whether the thing works. Training a policy is the easy half. The harder half is an evaluation harness that survives being interrupted, resets the scene on its own, logs every failure, and tells you honestly that you're at 92% and not at 100%. Right now I'm extending that work into Franka arms, sim-to-real transfer, and vision-language-action models for manipulation.

Experience

Instructional Student Assistant, Machine Learning · SJSU Department of Applied Data Science

Aug 2026 to present

  • Kept 40+ students moving through a graduate machine learning course, by grading assignments and exams and working with them directly on assignments, projects, and course logistics.

Graduate Research Assistant · SJSU Department of Applied Data Science

Mar 2026 to present

  • Scored 92% pick-and-place success and 2.49 cm mean placement error across 50 physical SO-101 rollouts, by deploying an Action Chunking Transformer policy against an evaluation protocol locked before testing began, with every trial tagged no-grasp, grasped-then-dropped, or wrong-placement.
  • Traced 100% of remaining failures to no-grasp perception rather than control, by debugging the full stack from camera to firmware: isolating 11 to 51 second policy stalls to out-of-distribution objects in view, and patching an STS3215 EEPROM re-lock defect that silently reverted motor IDs across power cycles and left the arm unusable.
  • Established that validation-loss checkpoint selection would have shipped a 67% worse policy (2.01 against 1.20 action MAE), by training 52M-parameter ACT and 99.9M-parameter SmolVLA policies in PyTorch/CUDA over 60K- and 20K-step runs and benchmarking 18 ms against 251 ms inference latency for closed-loop feasibility.
  • Published 49,633 demonstration frames across 50 episodes with zero integrity failures, by building a LeRobot v3 collection pipeline that synchronizes 6-DoF joint states and actions with dual 640x480 RGB streams at 30 FPS behind pre-training gates on frame alignment, video decodability, and NaN-free tensors.
  • Let operators run a 100-trial evaluation protocol unattended, by building a resumable harness with automatic return-to-start, stale-cache guards, video-measured placement error, and per-trial success and failure logging.

Software Engineer · Sapaad

Jun 2023 to Jun 2024

  • Served tenant-facing analytics on 5M+ rows a day, by shipping production Python ETL and ELT pipelines in Databricks and PySpark that extracted recommendation-model features across a multi-tenant platform.
  • Cut p95 dashboard query latency 43%, from 12.4 s to 7.1 s across 10+ Spark jobs serving 75+ tenant dashboards, by profiling execution plans, rewriting SQL window functions, and placing targeted caches.
  • Caught 14 schema, freshness, and ingestion anomalies automatically, by building a Delta Lake data-quality framework with schema validation and freshness checks across every ingestion pipeline.

Computer Vision Engineer Intern · Toothlens

Sep to Dec 2022

  • Removed the clinic's dependency on paired before/after patient captures it had no practical way to collect, by building an unpaired image-to-image translation pipeline in PyTorch for orthodontic preview generation.
  • Turned architecture selection into a measured decision rather than a visual one, by defining a cycle-reconstruction SSIM protocol that compared candidates under partial occlusion, low light, and off-angle capture, then choosing CycleGAN over Pix2Pix and variational autoencoder baselines and carrying it through to a working preview prototype.
  • Increased usable patient-image inputs 20% across 1,500+ real-world photos, by implementing OpenCV preprocessing guardrails that screened and corrected degraded captures before they reached the model.

Research Assistant · Illinois Institute of Technology

Sep 2019 to Mar 2020

  • Identified water-related disaster events from live Twitter data, by automating collection through the Tweepy API and generating FastText embeddings over the harvested corpus.
  • Reached 85% classification accuracy, a further 7% over the untuned baseline, by training an SVM over those embeddings and tuning it with GridSearchCV.

Education

M.S. Applied Data Science · San José State University

Expected Dec 2026

Coursework: Big Data Warehousing · Distributed Systems · Machine Learning · Deep Learning · Generative Models & Applications

AI & Machine Learning Operations Graduate Certificate · Indian Institute of Science (IISc)

B.S. Artificial Intelligence · Illinois Institute of Technology

May 2023

Coursework: Data Structures & Algorithms · Operating Systems

Skills

Machine learning & robotics
PyTorch · CUDA · scikit-learn · pandas · LeRobot · ACT · SmolVLA · Imitation learning · Teleoperation · NVIDIA Isaac Sim · MuJoCo · ROS 2 · OpenCV · Weights & Biases · Hugging Face Hub
Data & pipelines
Python · SQL · PySpark · Databricks · Spark SQL · Delta Lake · Parquet · Airflow · dbt · Snowflake · Kafka
LLM applications & retrieval
LangGraph · LangChain · Llama 3.3 · Llama 3.2 Vision · Groq · Pinecone · ChromaDB · Sentence Transformers
Services & delivery
Docker · FastAPI · GitHub Actions · AWS EC2 · React · TypeScript
Also worked with
JavaScript · SwiftUI · Vite · Chrome MV3 · Gemini Nano · Cloud Firestore · MongoDB · Redis · MySQL · Streamlit · PyTorch Geometric · RoBERTa · Grad-CAM · Cross-attention fusion · Mel spectrograms · Fine-tuning · Retrieval-augmented generation · Gazebo · CycleGAN · Pix2Pix · Variational autoencoders · SSIM

Projects

The rest of the work is on the home page, with the code and data behind each one.

Teaching & community

AI/ML Club, San José State University

Ran a hands-on LangChain workshop series teaching students to build applications on top of language models: prompts, chains, and putting the pieces together. Workshop materials

Robotics Club, San José State University

On the team building a rover for the University Rover Challenge, a Mars-analogue competition run in the Utah desert where student-built rovers have to do real work on rough terrain: retrieve and deliver equipment, run field science, and navigate autonomously between markers.

Away from the screen

Most of my spare time goes to a trail, a badminton court, or the gym. Cricket when there are enough of us for a side. And a sudoku when I want a problem that's guaranteed to have exactly one answer, which is not a guarantee the rest of my work offers.