---

I spent two years wrestling with the complexities of the Twitch video system and doing my best to contribute useful services where I can. Below are some brief summaries of the projects I worked on. Several details are omitted due to non-disclosure agreement.

In order to increase the redundancy of a service (details omitted), I had to port its distributed locking system from a third party service to AWS DynamoDB so that multiple entities could view the lock state of other entities around them. This required a deep understanding of how the system's locking actually worked, and writing rigorous tests to ensure that the overhaul was actually correct. To use a good phrase from a mentor of mine, I felt that this project bizarrely validated most traditional computer science undergraduate curricula.

My first project as a junior engineer was to build a service that detects abrupt drops in certain arenas of the video system's traffic. These drops are colloquially referred to as 'black holes', and the service is somewhat cheekily called LIGO. LIGO runs on AWS ECS/Fargate and connects to SQS, Kinesis, and Apache Druid. I architected, coded, and productionized the system entirely on my own, and gained a lot of experience on the journey.

Unequal cost multipathing, or UCMP, allows more efficient allocation of network traffic across the video system. I updated the previous algorithm (which implemented equal cost multipathing, UCMP's more inefficent predecessor) and worked with a colleague on the networking team to test the code live on the network. This project was yet another step in my journey to understand how exactly the internet works, something that I likely will never understand.

As an intern in the summer of 2020, I was tasked with building a service to track the Twitch video system's usage of third party CDNs which output a graph showing how many users were routed to CDNs per minute. I got my first taste of assembling AWS services into a single working program, and utilized Kinesis, Lambda, and CloudWatch.

---

Kids growing up in the age of AI need accessible online platforms with which to learn advanced concepts that will broaden their understanding of the computerized ecosystem they live in. The goal of GenAI was to design such a platform for learners of all levels. We intended to create a prototype to hand off to professional designers.

The platform GenAI was intended to be an online textbook with several chapters about different AI topics, such as neural networks, conversational agents, and robotics, among several others. The chapter we decided to prototype with was the one about generative adversarial networks (GANs) due to the project lead's personal interest in them.

A great deal of time was spent on the interaction design of the website itself, and keeping both the landing pages and the learning material itself as easy to navigate as possible for learners and users of all ages and levels. For example, we determined that while the color scheme and design should remain the same throughout the website, the "feel" (font, UI) should change between content levels.

We also wanted to make AI seem approachable, not futuristic, dark, chrome colored, or hacker-y. We considered naturalistic themes and light colors in opposition to the typical depiction of AI in media.

At the beginning of the project, I worked along with another student in charge of web design to wireframe out the look and feel of the website. I went through several of the educational websites that I used on a daily basis and tried to pick out the elements that I found the most natural and usable, and then combine these into a design. The web designer took my fairly simple ideas and gave them a great deal of color.

Halfway through the project, however, the web designer got too busy, and I was left with coming up with the visual design elements of the website as well. In accordance with our naturalistic ideas for representing AI, I chose a color palette, designed a v1 logo, and created a moodboard.

I was also assigned to write the beginner and intermediate levels of a GANs textbook chapter. This was the first time that I had to think deeply about what it meant to be a learner at each of these levels, and be mindful to use appropriate terminology and complexity. Similarly, we designed the visuals to "level up" as well.

This project taught me a great deal about designing UI and thinking very deeply about my user. Though I had never dabbled in visual design before, I enjoyed the chance to determine how the site looked and felt, skills that proved extremely useful to designing this website!

Close

---

Through my work with the MIT AI Education initiative, I've had a hand in creating a variety of websites for hosting our curricula and educational material. The sites are widely trafficked by educators worldwide.

DoodleBot

DoodleBot was the first site I built mostly on my own. That is where I learned core HTML and github.io deployment, which became my base for web projects.

aieducation.mit.edu

The MIT AI Education website is probably the most public facing and highly trafficked website that I'll work on for a long time. It houses the entirety of the MIT AI Education Initiative's resources, projects, publications, and fully tested curricula. This article that I helped write explains the initiative and its goals.

Creating this website was where I really honed my HTML and CSS skills, and learned how to run Javascript for interactivity as well. It's also the site that I'm the most emotionally attached, having committed 100+ times to the repo; my grad student and I were reluctant to let it be externally managed.

Dancing with AI

I'm mostly responsible for maintaining this site, but it's notable for being one of my first exposures to using the Jekyll framework, which came in handy for building this website! It was also my first experience with having educators contact me personally and ask for clarifications and fixes on curriculum material.

On their own, these websites were fairly small projects in the scope of things. However, tooling away on them gave me a much better feel for web dev and UI design that is coming in useful for this portfolio website in particular. The increase in my HTML/CSS skills has been, frankly, astronomical compared to where I started from. I used to find it insufferably frustrating, but now, working on these websites induces a sense of comfort and satisfaction in me similar to doing my laundry or cleaning my room. That feels like a huge improvement.

Close

---

When creating a game as huge as God of War, designers do an almost unimaginable amount of tweaking and twiddling with different game engine properties to get the gameplay feeling just right. My task as an intern was to create a prototype for an updated engine debugging menu. Several details are omitted due to non-disclosure agreement.

Like many triple A game studios, Santa Monica Studio utilizes a custom built engine for God of War. As such, the tool for tweaking settings within said engine was also custom designed, and was over a decade old when I was tasked with creating a more modern prototype.

I was tasked with using the Dear ImGui library to create this prototype. ImGui is an open source C++ library for creating user interfaces, primarily for game engines and other real-time 3D applications.

This was my first time working with a codebase on a massive scale. I learned how to parse through huge chunks of relatively undocumented code, and gained an intuition for how to look for the code actually relevant to my needs. I had also never worked with an external library as big as ImGui, so figuring out how to integrate it into the existing PlayStation codebase was a slow task.

I also learned what it was like to have multiple stakeholders for a system. Designers, artists, and programmers all used the engine debugging tool in different ways, and had different opinions on the changes that should be made to it. From the way the PS4 controller was used to navigate the menu, to the actual widgets used for certain features themselves, I learned the importance of taking multiple voices into account when creating an end product for a variety of users.

The biggest challenge I faced was struggling through the nuances of C++. I learned that low-level programming is definitely not my forte, and the intricacies of buffer overflows and overwritten caches did not bring me much joy. But in the end, the experience of working in a triple A studio on such a huge project was worth all the frustration. As a small contributor, I'm excited for the world to see God of War: Ragnarok, and as a fan of the franchise, I'm dying to play it myself!

Close

---

TaleBlazer is an augmented reality (AR) software platform developed by the MIT Scheller Teacher Education Program (STEP) Lab that allows users to play and make their own location-based mobile games in a browser-based block programming language. My task as an undergrad researcher was to prototype and add certain highly requested blocks to the web editor.

An important feature for AR games, which happen both in real locations and in real-time - is time! When I joined the TaleBlazer project, some of the most highly requested features from users were time-based event blocks, in which a user could trigger an event using elapsed time. My task was to brainstorm, prototype, and implement the best set of blocks for users to achieve this functionality.

I began with brainstorming a variety of time-based features that could come in handy in a location-based AR game:

I then created paper prototypes and iterated upon them multiple times before finally moving on to implementation. The TaleBlazer backend is written in Javascript on top of a CakePHP backend. I primarily wrote Javascript in Android Studio, and built countless iterations of the app onto an old Android phone. The final result was that these three blocks were added to the TaleBlazer programming language:

My project for TaleBlazer was, in itself, relatively simple. However, this summer was absolutely instrumental to my growth as an engineer for several other reasons. I learned how to use Git, and managed, committed to, and pull requested branches with my chcanges. I learned the process of prototyping and adding user requested features, and learned how to go through several iterations of ideation. I learned how to adjust to different development environments, and code in languages that I had never encountered before. For all of these reasons, my summer at TaleBlazer will likely remain the most formative summer of my career.

Close

---

From September 2017 to January 2021, I wrote 77 blog posts (an auspicious number in retrospect, given how much I love Final Fantasy VII) for the MIT Admissions website. The mission of the Admissions blogs is to paint a genuine, unedited picture of MIT life and culture to potential applicants.

From September 2017 to January 2021, the MIT Admissions front page got just about 40 million pageviews with around 7 million unique users from all around the world. With my blog posts, I contributed to a small amount of those pageviews with a variety of different types of blog posts: compilations of MIT culture, several enthusiastic posts about Spotify, and some minorly viral data visualizations. Compiling most of these posts helped me develop what I now consider to be a life skill: shamelessly asking large groups of people to provide me with similarly large amounts of data. At some point, I even had a salty MIT Confessions post written about just how shameless I was. However, I firmly believe that these experiences built character, and will serve me well in the working world. When I need to get something done, I'll get it done.

I had never written for such a large audience before, and it was both strange and gratifying to receive comments and emails from incoming students telling me how much my blogs resonated with them on different levels. Being told that my blogs were the reason that a student applied to MIT certainly ranks among the more touching experiences in my life. In the grand scheme of things, the impact of my blogs on the internet was quite small, but if they could change even a few people's life trajectories, then I think my quest as a writer has been fulfilled.

Close

---

I have been into bikes since Jaipur days. First ride was a Hero Glamour 125cc. Not a beast, but pure vibe. I learned basic wrenching there: carb, plugs, engine tone, all that garage stuff.

Then came Ignitor. Better pull, smoother shifts, and many Jaipur to Pilani runs. That machine got stolen later, still hurts.

Now in Singapore I ride a Honda CB190X. License cap is 200cc, so this is my max-spec legal setup here. For me it is perfect: upright ergonomics, stable suspension, clean throttle response, and chill long-run comfort. Late-night empty roads are elite therapy mode.

Sometimes I record rides too. Sample below.

---

Music has been a long side quest. I started with harmonium and vocals as a kid, paused for a while, then came back hard with keys and guitar in college. Mostly self-driven grind, jam-heavy learning, and random stage gigs.

Performances (Public and Personal)

---

Teaching keeps my fundamentals sharp. My flow is simple: concept map first, equations next, then problem-solving reps till the logic sticks.

My Teaching Adventures

• Fall 2021: Mechanisms and Machines, BITS Pilani
  TA for this course at BITS Pilani. Recordings got lost, but that semester taught me how to explain tough mechanics in plain language.
• 2021-2022: AutoCAD Sessions, BITS Pilani
  Led some AutoCAD workshops. Majorly as a part of the M&M course.
• Fall 2024: Fluid Mechanics, NTU Singapore
  Took in-person sessions at NTU with interactive discussions and real-world case breakdowns.
• Spring 2025: Heat Transfer, NTU Singapore
  Ran Heat Transfer classes at NTU with a problem-first style and concept clarity focus.
• Fall 2025: Heat Transfer (Round Two), NTU Singapore
  Back with Heat Transfer at NTU. Going all in to make it a party. Check the YouTube playlist for the Fall 2025 tutorials!

Fall 2025 Heat Transfer Playlist

Lecture Notes

More notes and tutorial drops coming every semester.

Close

---

I was a design collaborator for the Science+C initiative through the Educational Development Center. I designed a three day data science curriculum for integration into high school biology, chemistry, and physics classes. I ran workshops with educators and iterated using their feedback. The curriculum will be used by the State of Massachusetts.

Science+C is a collaboration with the Massachusetts Department of Elementary and Secondary Education. The goal is simple: give high school students a practical path into computational science. The three courses are Computational Biology, Computational Chemistry, and Computational Physics, and they are designed for statewide use.

I specifically worked as a design collaborator on the Data Science + Machine Learning units. The units were divided into data science, decision trees, clustering algorithms, and neural networks. I designed the data science curriculum, a 3 day workshop with a lesson and associated activities for each day.

I created lesson plans designed to introduce beginners to data science at a low level. I cover the basics of data science and what data scientists to, and designed activities to teach summary statistics and visualization usage in an interactive manner.

I ran this curriculum with high school science educators, then iterated using their feedback. That cycle helped us simplify the lessons and improve student flow.

The curriculum will continue to go through iteration cycles until it pilots with schools across the state of Massachusetts in September 2022.

Close

---

After about a year working at Twitch, I decided that it was time to test out the service that I work on! I'm starting with streaming my very favorite game - Final Fantasy VII, if that wasn't obvious from the rest of this site - and I'm hoping to stream more of the classic Final Fantasies, playing guitar, and maybe - just maybe - myself applying to grad school. Reaching Partner status seems like a lofty goal, but I'd like to at least reach Affiliate.

Check my channel out at W0LFHEART on Twitch!

---

At the University of Tokyo research camp (July 2025), I worked in a 46-member global cohort from 10 universities. Mission: build a practical, high-throughput plastic sorting system for real recycling lines.

Our architecture combines electrostatic separation, acoustic resonance, and electromagnetic control. It hit 90-95% sorting accuracy on mixed plastic streams and stayed stable even with messy input feed.

How it works:

1. Electrostatic Separation: Plastics get triboelectrically charged as they slide along the belt, then pass through a high-voltage electric field (~40 kV) where charge differences cause lateral deflection, achieving >95% sorting accuracy.

2. Acoustic Resonance Separation: Using ~40 Hz acoustic waves, plastics vibrate differently based on their stiffness (Young's modulus), allowing separation with ~90% accuracy through resonance amplitude variations.

3. Electromagnetic (EM) Separation: The breakthrough component. Pulsed EM fields at 60 kHz and 100 kHz target differential dielectric polarization dynamics. PVC, with chlorine atoms in its polymer chain, has higher dielectric loss and sticks to the belt longer at 100 kHz, while PET slides off, enabling >95% separation accuracy.

My primary contribution focused on the physics and system modeling for the electromagnetic module, tuning the pulsed field frequencies and stick-slip parameters to cleanly separate PET and PVC. I coordinated with external APAC collaborators to validate our models and developed comprehensive testing plans for handling dielectric polarization with ionic dynamics on a conveyor belt.

Impact and Specifications: Our prototype processes 50 to 1,500 kg/hour depending on setup, with estimated cost around US$118k to 195k including maintenance. As global plastic use grows, this system can reduce contamination, improve sorting quality, and keep industrial throughput practical.

After weeks of equations, design reviews, and rapid prototyping, we shipped a working system with strong reliability and practical complexity. Not just a demo. It is a serious step toward circular manufacturing.

NTU Team Experience: Special thanks to Prof. Yang Jianfei, who arranged this incredible opportunity for us NTU students and captured many memorable moments throughout the camp.

This project won Best Innovation and gave us a clean blueprint for scaling smarter sorting systems.

Close

---

This work is about cold-chain reliability for vaccines and food-grade payloads. Core stack: thermal modeling + IoT telemetry + ML forecasting. Goal: detect risk early and prevent thermal excursions before quality drops.

CRYOMOVE: Cold-Chain Prediction Stack

Centerpiece is CRYOMOVE, an ML-driven prediction stack for PCM behavior in real time. This is built for proactive ops decisions, not passive dashboard watching.

How CRYOMOVE Works:

• AI-Powered Predictions: An Artificial Neural Network, trained on extensive ANSYS Fluent simulation data, accurately predicts PCM melt times based on live temperature readings
• Real-Time Monitoring: Continuous temperature sensing provides instant feedback on cold chain status
• Proactive Interventions: The system enables preventive action before temperature excursions occur, ensuring vaccine integrity
• Cloud-Based Intelligence: Deployed on cloud platforms for scalable, accessible monitoring and dynamic decision-making
• Versatile Applications: Beyond vaccines, the system adapts to food preservation and other perishable goods across industries

Experimental Progress

We run full simulation-to-test loops. Not just CFD figures. Actual box builds, material checks, and stress-case validation.

Current Achievements:

• Extended Performance: Successfully extended cold box performance to 6-7 days, nearly doubling the industry standard of 3-4 days
• Advanced Materials Testing: Evaluating insulation foils and thermal paints for stronger thermal protection
• Environmental Control: Investigating the effects of humidifiers and desiccants on cold box performance
• Glycerol-Modified PCMs: Developing enhanced phase change materials for superior thermal regulation

Next Build Phase

Next phase is stronger forecasting and robust deployment under noisy field conditions.

Advanced Neural Network Development:

• Physics-Informed Neural Networks (PINNs): Integrating fundamental physical laws with machine learning for unprecedented accuracy
• Recurrent Neural Networks (RNNs): Enhanced time-series forecasting for complex temperature patterns
• Real-World Training Data: Incorporating diverse, realistic temperature profiles from various environmental conditions

Dynamic Environmental Integration:

The ultimate goal is a fully integrated system that considers:

• Real-time traffic data affecting transport times
• Weather conditions impacting thermal loads
• Geographic mapping for route optimization
• Dynamic environmental factors for predictive performance modeling

Why This Matters

Impact is direct: better vaccine potency retention, safer medicine delivery, and lower food spoilage. The stack is designed to scale from local clinics to full logistics networks.

In short, this is predictive thermal logistics with real ops value.

Close

---

I use PIV pipelines with Vicsek-model thinking and cross-correlation tracking to study collective motion. Focus areas are flow field reconstruction, velocity coherence, and emergent dynamics in biological swarms.

Close

---

This work targets high-performance cooling in mini-channel systems for solar, EV thermal management, and industrial exchangers. The toolbox includes magnetic nanofluids, vibration control, and external magnetic forcing.

We study how vibration frequency and magnetic field strength reshape boundary layers and secondary flows. Target is higher heat transfer with controlled pressure penalty. Results are published in Results in Engineering and Numerical Heat Transfer.

We also run exergy analysis to track energy quality, not only temperature reduction. End goal is robust thermal design that survives real operating constraints.

• Timeline: January 2022 - Present
• Mentors: Dr. Suvanjan Bhattacharyya
• Collaborators: Naman Jain, Ali Cemal Benim

Close