Software Development Engineer Intern (ML)
Jun 2026 — Aug 2026ML EngineeringInternship
Built self-serve ML training infrastructure on AWS for internal robotics teams.
PythonAWSML Systems
Ishan Dave
Experience
Industry and Research Work
Industry
3Software Engineering Intern
May 2025 — Aug 2025Full-StackCloud
Sped up backend services with C++ modules and consolidated real-time AI demos onto one React + FastAPI platform on AWS.
ReactFastAPIAWSC++
Software Engineering Intern
May 2024 — Aug 2024Full-StackInternship
Built self-service registration and a Japanese/English i18n layer for an AI sensing platform.
FlaskSQLAlchemyi18nJavaScript
Research
3Lead ML Researcher
Aug 2025 — PresentResearchMLBiotech
Designing LLM-driven search and scoring methods for RNA therapeutic discovery.
PythonLLMsOptimizationResearch
AI Research Intern
Jan 2026 — May 2026ResearchNLPTrustworthy AI
Built a hallucination-detection benchmark and an NLI classifier to evaluate it.
PythonBERTNLIEvaluation
LLM Research Intern
Sep 2025 — Dec 2025ResearchVLMsInterpretability
Studied where vision-language models stop using visual input and where reliability signals live.
PythonPyTorchVLMsEvaluation
Publications
Peer-reviewed research
ICML 2026
Where Reliability Lives in Vision-Language Models: A Mechanistic Study of Attention, Hidden States, and Causal Circuits
Logan Mann, Ishan Dave, Ajit Saravanan, et al.
A mechanistic study of where reliability is encoded in VLMs, across attention, hidden states, and causal circuits.
Read on arXiv↗
ICLR 2026
Visuals Lie, Consistency Speaks: Disentangling Spatial Attention from Reliability in Vision-Language Models
Logan Mann, Ishan Dave, Yi Xia, Ajit Saravanan, Saadullah Ismail, Shikhar Shiromani, Emily Huang, Ruizhe Li, Kevin Zhu
Spatial attention in VLMs is uncorrelated with accuracy; reliability lives in late-layer hidden states and generation consistency.
Read on arXiv↗
3rd @ SCVSEFA
Automated Detection of Parkinson's Disease Patterns in Spiral Drawings Using Convolutional Neural Networks
Ishan Dave
A CNN approach to Parkinson's detection from digitized spiral drawings, as a non-invasive diagnostic alternative.
Read Paper↗
Projects
Selected personal and academic builds
IceBreaker
Academic Outreach Platform
Automated professor outreach for 40+ students with FastAPI, SQL, and Gmail API integration on Linux.
FastAPISQLGmail APILinux
$
Aionix
Founder · Tech-Literacy Platform for Seniors · 2023 — 2025
Full-stack learning platform for senior citizens, with automated CI/CD releases to AWS.
FastAPIAWSSQLiteCI/CD
$
Current Focus
ML & AI Research
Leading ML and search work for RNA therapeutic design at Berkeley's Sky Computing Lab, alongside research on vision-language model interpretability.
PyTorchLLMsVLMsResearch
ML Engineering & Software
Taking models from research to production — most recently building ML training infrastructure at Amazon Lab126.
PythonAWSC++Systems
Physical AI
Drawn to the sensing, dynamics, and control problems of real-world deployment, and to closing the sim-to-real gap.
RoboticsControlEmbodied AI
Technical Stack
Languages
- Python
- C
- C++
- Java
- JavaScript
- SQL / NoSQL
ML / AI
- PyTorch
- TensorFlow
- HuggingFace
- QLoRA / PEFT
- BERT / NLI
Web / Tools
- React
- FastAPI
- Flask
- SQLAlchemy
- HTML/CSS
Cloud / Infra
- AWS (SageMaker, S3, EC2, ECR, Lambda)
- Docker
- GitHub Actions CI/CD
- Git
- Linux
Education
UC Berkeley
Computer Science + Applied Math
Expected graduation: May 2028
What I've been taking
Beyond Classes
Nova Consulting
Project Manager
Berkeley technical and strategy consulting—client-facing projects across software and analytics.
Rec Basketball
Sharpshooter
Sharpshooter at the rec during my free evenings.