Dharun Narayanan
Software Engineer at Microsoft
I build AI experiences for everyday work, agentic tools for engineering teams, secure systems that hold up, full stack products end to end, and things people actually like using.
Good engineering is invisible. Bad engineering is somebody's bad day.Bad engineering is somebody's bad day.
About
I design and ship systems where the engineering is rigorous and the experience is humane.
Builds as Software Engineer · AI Engineer · Machine Learning Engineer · Security Researcher · Full Stack Developer · Human Centred Builder
The thread through all of it is human centred design: whatever I build, the person on the other side of the screen comes first.
Based in Bellevue, Washington
Education
1 of 3
M.S. Computer Science
Oregon State University, USA
2023 to 2025
GPA 3.93 / 4.00
B.Tech Computer Science
Amrita Vishwa Vidyapeetham, India
2019 to 2023
GPA 3.69 / 4.00
Higher Secondary
Rasi International School, India
2017 to 2019
89% aggregate
What I work with
Three disciplines that only hold together as a set. Choose a group and its ring comes forward.
Languages
- Python, core skill
- TypeScript, core skill
- Java, core skill
- JavaScript
- C / C++
- SQL
- Dart
Frameworks and Web
- React, core skill
- Node.js
- Flutter
- HTML
- CSS
- REST APIs
AI and Machine Learning
- PyTorch, core skill
- Deep Learning, core skill
- NLP
- Computer Vision
- scikit learn
- Pandas
Security
- Security Engineering, core skill
- Vulnerability Analysis
- Intrusion Detection
- Offensive Security
- Threat Modelling
Cloud and DevOps
- Azure, core skill
- AWS, core skill
- DynamoDB
- SQS
- Docker
- Ansible
- Firebase
- Git
Craft and Practice
- Human Centred Design, core skill
- UX Research
- Accessibility
- Technical Writing
- Mentoring
At Microsoft
I work on the Copilot experience in OneDrive and SharePoint, and on how my team builds it: an agentic harness that takes a change from an idea to a pull request.
- Spec, with an agentAn engineer and an agent shape the change together.
- ApproveThe engineer reads the spec and signs off.
- ImplementAgents build it in the monorepo, with the tools it takes.
- Pull requestThe work arrives as a pull request, ready for review.
Under the hood the harness runs investigations across several agents at once, orchestrates them, and adjudicates the results at every stage. Each hand off goes to the model class best suited to it.
Where I have built things
Eight roles across industry, research, teaching and security operations, newest first.
Feb 2026 to PresentCurrent
Sept 2024 to Dec 2025
Graduate Research Assistant
Oregon State University
Researched provenance based intrusion detection, reconstructing system level causality graphs to surface the intrusions that signature matching misses.
Jun 2025 to Sept 2025
Software Engineer Intern
Microsoft
Delivered an online evaluation framework for the autofill columns feature in SharePoint Online, using a large language model to categorise columns at scale, and built the interactive internal dashboards the team used to read user engagement.
Jan 2025 to Apr 2025
Software Development Engineer Intern
Amazon
Built an internal tool, surfaced through Amazon's front end MVC framework, that shows verification status for VetDiet pet prescription orders.
Jan 2024 to Dec 2024
Security Analyst / Security Engineer
ORTSOC, Oregon State University
Analysed vulnerabilities, hardened security infrastructure and ran assessments for the university's regional security operations centre.
Mar 2024 to Jun 2024
Graduate Teaching Assistant
Oregon State University
Supported instruction, grading and student mentoring across computer science coursework.
Jan 2023 to May 2023
Undergraduate Student Researcher
Amrita Vishwa Vidyapeetham
Researched embedded systems and applied machine learning.
Aug 2021
Tech Mentor
The Shooting Star Foundation, California
Mentored students in technology fundamentals.
View certificate (opens in a new tab)
Projects
Twelve projects across AI and machine learning, security and full stack.
Human centred design
Beaver Rides
Will I make it to class?
A ground up redesign of Oregon State's shuttle app. User research showed that was the one question riders had, so I rebuilt the information architecture around answering it, not around the transit data model.
Source Beaver Rides (opens in a new tab)
NLP
MELOSCRIBE
Music a human pianist can actually play.
Turns plain language descriptions into music and piano notation. Generating notes was the easy part; the real build was constraining the output to what real hands and real notation allow.
Source MELOSCRIBE (opens in a new tab)
Deep learning
Brain Tumour Segmentation (BraTS)
Not every mistake costs the same.
A segmentation pipeline that finds tumour regions in multimodal MRI volumes, built knowing that a missed tumour weighs far more than a false alarm.
Notebook Brain Tumour Segmentation (BraTS) (opens in a new tab)
Web
E-Library
Browse, borrow, return.
A department library system with role based access for staff and students.
Live demo E-Library (opens in a new tab)
Web
Hungrezy
Dinner, tracked.
A food ordering service with restaurant browsing and a live cart.
Live demo Hungrezy (opens in a new tab)
Web
ESMS Web
Conflicts solved before students see them.
An exam scheduling platform that matches rooms, invigilators and timetables.
Live demo ESMS Web (opens in a new tab)
Mobile
ESMS Mobile
Your exams, on the device you check.
The companion app to ESMS Web, with each student's exam schedule and a notification when it changes.
Source ESMS Mobile (opens in a new tab)
NLP
Code-Mixed Translation
How multilingual people actually write.
Normalises code mixed text such as Hinglish and Tanglish into a single target language.
Source Code-Mixed Translation (opens in a new tab)
Computer vision
U-Net Segmentation
A controlled testbed for encoders and decoders.
Shape outline segmentation built on the UNet architecture.
Notebook U-Net Segmentation (opens in a new tab)
Machine learning
Forex Predictive Model
Honest backtesting.
A model of forex price movement that makes no curve fitted accuracy claims.
Notebook Forex Predictive Model (opens in a new tab)
Machine learning
Crop Prediction System
Built for low connectivity.
A Naive Bayes recommender that suggests suitable crops from soil and climate inputs.
Source Crop Prediction System (opens in a new tab)
Freelance
Phantom Studios
Brief, build, handover.
A marketing site for a nontechnical owner.
Live demo Phantom Studios (opens in a new tab)
Let's build something.
I'm always up for interesting conversations: research collaborations, thorny engineering problems, or just comparing notes. If you've got something you think I'd find interesting, I probably will.
Find me
Write to me
Message sent
Thanks, your message is on its way. I'll get back to you soon.