Weeks 23 - 26
Spring 2026
One class that I have been enjoying a lot this semester is the Information Science seminar series. The format is simple: each week, a researcher from a completely different field comes in and talks about how they think about technology, data, and AI.
Something that makes this class interesting is how different the perspectives are. Some speakers approach technology through law. Some through religion. Some through political science or environmental justice. AI is everywhere. It's embedded in institutions, cultures, and social structures.
Here are a few talks that stood out in February:
On Judgment - A Critical Grammar for Computing and Law
One idea from this talk that really stayed with me was the idea of tractability. In computing, we often assume problems are well-defined and solvable if we just design the right algorithm. But Gerardo Con Diaz pointed out that tractability is actually something we create by deciding what counts as measurable, what ambiguity we ignore, and who has the authority to decide. Law does this through procedure and precedent. Computing does it through formal logic and machine-readable rules. It made me think about how every ML pipeline or system architecture encodes decisions about what the system can “see” and what it cannot. In other words, building software is also building a framework for judgment.
Religion and AI: Theoretical and Empirical Approaches
This talk pushed me to think about AI outside the usual engineering mindset. Nesrine Mansour explored how religious and philosophical traditions interpret emerging AI systems - questions about agency, moral responsibility, and the role of machines in human decision-making. What I took away is that technologies like AI become part of cultural and ethical systems. As engineers or data scientists, we often focus on performance metrics, but conversations about AI increasingly involve deeper questions about meaning, authority, and values. The impact of AI systems will always extend beyond the codebase.
Ethnicization of Conflict
This talk was particularly interesting from a data science perspective. Alexandra Siegel studies political behavior using massive datasets of social media posts and applies text analysis, network analysis, and machine learning to understand how narratives about identity and conflict spread online. We saw how computational tools can reveal patterns in political discourse that are impossible to observe manually. At the same time, these systems also shape the very environments they analyze. It reinforced how data science sits at the intersection of technology and society - analyzing digital systems while also recognizing that those systems influence real-world behavior.
Frontline Futures - Advancing Environmental and Climate Justice
Katlyn Turner’s talk focused on how technological infrastructures - from AI systems to energy supply chains - affect communities that are often excluded from the decision-making process. Her work highlighted how innovation can unintentionally reinforce social hierarchies if equity is not considered during design. What I found particularly interesting was her idea that technological systems always produce externalities, and those costs are rarely distributed evenly. Building systems responsibly means thinking about the broader context in which those systems operate.
AI Builders Boulder
One evening this month I biked about seven miles to attend an AI Builders Boulder meetup, and it was absolutely worth it. The room was full of engineers and founders building real AI products, with teams from BranchLab, Wabi, Freeplay, and Slider sharing what they’re working on. The conversation around AI has shifted. The question is no longer “Can we build a model?” - that part is largely solved. The real challenges now are around evaluation, deployment, monitoring, and reliability in production systems.
A few themes kept coming up throughout the talks. Evaluation is becoming its own discipline. Tooling and infrastructure matter just as much as modeling itself, and the real differentiation between companies often lies in how they design systems around their models. Hearing this was especially interesting given the work I’ve been doing with LLM extraction pipelines and confidence scoring systems, where reliability and validation become critical once a model is integrated into a workflow. Modern AI work is increasingly about building robust systems.
Graduate Alumni Connect
This was an in-person event where current students could talk with CU Boulder alumni working in industry. It turned into a very engaging evening, especially hearing from alumni working in tech at companies like Lumen and Gravity. These were really honest people, and so were the conversations. People spoke openly about career paths, how quickly AI is evolving, and how important it is to keep learning outside of formal coursework.
 |
| (Found me?) |
One piece of advice that stuck with me was surprisingly simple: “Would they want to have a beer with you?” In other words, technical skills get you in the room, but people skills determine how far you go. The alumni talked a lot about networking, staying curious, and actively putting yourself in environments where interesting conversations happen.
That resonated with me because I’ve been trying to do exactly that. Over the past year, I’ve gone out of my way to attend events whenever I can - taking buses to Denver for Colorado Startup Week, biking across Boulder for meetups, sometimes in freezing weather. It might feel like small effort at the moment, but these small efforts compound. Being in rooms with builders, researchers, and alumni is often where the most valuable learning happens.
Amply - Energy Optimization Simulation
This month I also worked on the first milestone of Amply, an energy optimization simulation that explores how compute workloads could make smarter decisions about when to draw power from the grid versus a battery system. The prototype pulls real hourly grid demand data from the EIA API and uses it as a proxy for price signals. On top of that, I built a simulation layer that models battery behavior (charge thresholds and rates) alongside simulated GPU compute demand.
Okay, that was honestly a lot of electrical terms for someone who studied V = IR years ago. The core idea is simple: every hour the system decides whether it should charge the battery, run off the grid, or run off stored energy. Running the simulation over a 30-day period made it easier to see how energy demand patterns and battery constraints interact in practice.
What I found most interesting while building this was the decision logic problem. The first version uses a simple rule-based approach, but it quickly became clear how many factors could influence optimal behavior: time-of-day effects, weekday demand cycles, price volatility, and forecasting signals. It turned the project into a small systems design problem rather than just a modeling task. My next steps are to refine the decision engine, incorporate better price signals, and experiment with simple forecasting or optimization approaches while improving the visualizations to make decision states clearer. Check out the project here!
Data Science Alumni Panel
We also had a Data Science Alumni Panel this month featuring some impressive CU Boulder alumni: Sushil Deore (Product @ Samsung), Varad Luktuke (Data Scientist @ Deque Systems), Praveen Kumar Myakala (VP, Senior Lead Software Engineer @ JP Morgan Chase), and Sorabh Kaila (CFO @ TLCx). It was great hearing about their career journeys, how they navigated transitions across roles, and what skills actually matter once you’re out in the industry.
One piece of advice that came up repeatedly was the importance of networking for international students. It’s something I’ve now heard from many different people in different contexts, which probably means it’s worth paying attention to. At first, putting yourself out there at events can feel uncomfortable, and I've been there. But after attending a few meetups or panels, it becomes much more natural. Conversations start flowing, you learn what others are working on, and sometimes opportunities appear in places you wouldn’t expect. In a way, the same rule applies to both your career and your love life: GET OUT THERE!
Hackverse
I had the amazing opportunity to participate in an overnight 12-hour MS-DS Hackathon (Hackverse). We almost won. Almost. We missed it by 0.8 points, which still stings a little. There were three companies presenting three completely different problem statements, and in classic over-ambitious fashion, my team Code Blooded tried to work on all three.
 |
| my amazing team :) |
Personally, I was most excited about Ricoh’s challenge, which involved building an agentic AI support system that could reason across multiple technical manuals, retrieve relevant sections, and generate traceable troubleshooting instructions with citations. It was very close to the kind of systems thinking I’ve been exploring recently. Unfortunately, we ran out of time before we could finish the full prototype. In hindsight, skipping that two-hour power nap might have helped… but honestly, the learning was worth it.
My teammates also worked on two other interesting challenges. LexTrack AI focused on automatically identifying relevant Reddit communities for niche products and generating subreddit-specific outreach posts. ClinsightAI explored extracting operational insights from hospital review datasets to help healthcare organizations understand the drivers behind patient ratings and operational issues. Each problem pushed us to think about how AI systems move beyond simple analysis into decision intelligence and actionable insights.
And of course, it wouldn’t be a hackathon without the essentials: pizza, biriyani (for the first time ever in a hackathon), and slightly sleep-deprived engineering decisions.
Honestly, I wanted to finish building these projects before writing this blog. But life happens. I’ll definitely return to them later because they’re fantastic learning opportunities. Just maybe this time with a good night’s sleep instead of an all-nighter. Gone are the days when I could stay up all night and function without coffee.
Parsyl
My time at Parsyl is going by incredibly fast. It’s already been a little over two months since I joined, and the amount I’ve learned in that time has been unmatched. I know so much more today than I did when I first started. Working alongside people who are deeply thoughtful about data, risk, and systems has been an amazing learning experience, and I’m very grateful to be part of the team.
One highlight this month was getting the opportunity to lead one of the Data Science up-skilling sessions. I gave a 40-minute talk on different ways to measure confidence in LLM outputs, walking through approaches for evaluating reliability and uncertainty in generated responses. What made it particularly exciting was that I was presenting to senior staff across the company. Leading a technical discussion like that felt like a big milestone for me.
Something else I’ve really appreciated about Parsyl is how much the company values learning and experimentation. Teams actively make time to explore new tools, discuss AI capabilities, and think about how technology can improve workflows and processes. Being in an environment where curiosity and upskilling are encouraged makes a huge difference, and it’s been great to experience that culture firsthand.
V7 Demo
I also had the chance to attend a demo with V7 Labs, and it was very interesting to see how they’ve approached document extraction and classification. Their platform makes it surprisingly seamless to process large volumes of documents and turn unstructured information into structured data. I appreciate how much thoughtful tooling and interface design go into making complex ML pipelines usable in real-world settings. Building effective AI systems is about the infrastructure and tools that make those models reliable and easy to integrate into everyday workflows.
Research and Innovation Office - Impact Intern
Earlier this month, I had the opportunity to present the Pivot RP funding opportunities website I built to senior staff at the Research & Innovation Office at CU Boulder. It was great to hear their feedback on how useful a tool like this could be if integrated into the main RIO website. The system simplifies how funding opportunities are discovered and organized, improving accessibility for faculty while reducing the manual back-and-forth that typically happens when compiling and publishing these opportunities.
Later, I also had the chance to discuss the project and its potential next steps with Mike Mitchell, Senior Director of Research Development. We talked about how tools like this could make funding opportunities significantly easier for faculty to discover and explore. The platform essentially automates the process of collecting, cleaning, and publishing Pivot RP funding opportunities - something that previously required hours of manual effort. If integrated into the workflow, it could save several hours of administrative work each week while making funding information more accessible across campus.
Graduate Seminar Series – Fundamentals for Academic Success
Another session this month focused on something deceptively simple: how we manage our time and energy. We talked about concepts like time blocking, task batching, and time auditing. The interesting part was the idea that productivity is less about doing more and more about becoming aware of how you actually spend your time and energy. The discussion around barriers to action, spending vs. investing time, and identifying weak points in daily routines really stuck with me. Sometimes small changes in structure can make a big difference in how effectively we work.
The session also introduced the Enneagram framework, which looks at different personality types and the motivations that drive our behaviors. While no personality system is perfect, the value was in using it as a tool for reflection - asking why we approach work the way we do, what motivates us, and where our blind spots might be. Whether it’s in research, industry, or everyday work, understanding our own patterns can make a big difference in how we grow, collaborate, and solve problems.
Well, "A curious systems thinker who loves learning deeply, building things, and then sharing the work with the world" is what I was told I am. And it sounds about right.
MS DS Career Snapshot
Another interesting session this month focused on what actually worked for MS-DS alumni when landing jobs. Looking at real outcomes from recent graduates, the data showed that the strongest clustering of roles was in Machine Learning / AI Engineering and Core Data Science, followed by analytics-focused roles and leadership or strategic positions. It was helpful to see the job market through actual alumni outcomes rather than general industry advice. One takeaway for me was that many successful paths seem to combine strong technical foundations with applied systems work, especially in ML/AI engineering contexts. These patterns made the career landscape feel a lot more concrete and gave me a clearer sense of where different skills and projects can realistically lead after the program.
 |
| credits |
CU Wizards Show
This had absolutely nothing to do with data science - or maybe it did in spirit. I went to the CU Wizards science show without really knowing what to expect, and it was honestly amazing. For a moment it felt like going back to being a kid again, watching experiments and demonstrations that remind you why science is exciting in the first place. Sometimes in graduate school, it’s easy to get caught up in models, deadlines, and projects, but events like this are a good reminder of the simple idea that probably brought many of us into STEM in the first place: science is just really cool. (NERD!)
- I started meal-prepping. The hope is that I’ll eat more at home and stop buying food outside all the time. Also, to save time. LLMs are weirdly good at helping me scale recipes up and down, which has been surprisingly useful. At this rate, I’m basically on my way to becoming a chef.
- I also started learning Spanish. Maybe because of the Super Bowl. Maybe not. So far, it’s been pretty manageable. Duolingo does a great job of structuring lessons, so you keep repeating things until they’re etched in your brain. They’ve also added these cool interactive scenarios like “You’re at a restaurant, and you’re the waiter” or “You’re on a video call speaking Spanish.” Those are actually pretty fun. Also, the reminders when you’re about to lose your streak… intense. It’s a love-hate relationship.
- I went ice skating. No, not because of Alysa Liu. Being on ice is a weird feeling - like you probably won’t fall… but you also might. I’ve only gone once. We'll see if I ever try it again. But seriously, huge congrats to Alysa Liu. I’m not impressed just by the skating but by her mindset. Every second you’re out there, you’re learning something. Nothing feels impossible when you’re actually enjoying what you’re doing. Watching her skate, you could tell she was genuinely having fun.
- I’ve also started rock climbing. Yes, apparently I’m collecting hobbies now. I’m using the auto-belay for now and definitely need to build more upper-body strength. I’ll admit, there’s a bit of peer pressure involved because a lot of people around me love climbing. Not saying I dislike it - it’s just very new. But I’m definitely doing better than day one, and I’m glad I’m trying new things.
Comments
Post a Comment