Ashia Nagi

Excited to be mentoring, alongside channa weeratunge, a graduate-level software engineering capstone project this semester for Dr. Pradeep Buddaraju at the University of Houston-Clear Lake. Our students are building an Agentic AI application that researches market data and provides actionable recommendations for options traders focused on covered call strategies. It’s been refreshing to see the team quickly tackle requirements, domain research, and high-level design. You can follow their progress on the project site https://lnkd.in/dVNTpTxR A parallel goal of the project is to explore how AI can also contribute to building the application — using practices like test-driven development (TDD) to ensure contracts are honored throughout the SDLC.

113

5 Comments

Texas has suspended new H-1B visa filings at public universities following an order from Governor Greg Abbott. The freeze will remain in place until May 2027, unless institutions obtain special approval. The policy is expected to impact hiring across public universities, particularly affecting professionals, researchers, and faculty members who commonly rely on H-1B visas. Supporters say the measure is intended to safeguard local employment opportunities, while others express concern that it could limit access to international talent and affect research and academic competitiveness. #usimmigration #usa

3

My grad school research might decide which streets Austin fixes next. For my MS at UT Austin, I built a Machine Learning model to help Vision Zero prioritize cyclist safety infrastructure. I analyzed 2,757 crashes over 10 years, but with a twist: I "blinded" the model to demographics and only included factors the city can actually change (speed limits, intersection design, lighting, bike lanes, etc.). Key finding: Speed limit is 3x more predictive of severe injury than bike lane presence. I cold-emailed Joel Meyer, Austin's Transportation Safety Officer. His response: "Very impressive... we're definitely interested in approaches that allow for more proactive identification of risk factors." We're meeting next week to discuss how the city might use this. Sometimes the best thing you can do with research is send it to someone who can use it.

74

For the past few months, I’ve had the opportunity to work as a Research Assistant at the The University of Texas at Arlington under Dr. Genevieve Graaf. My focus has been a deep dive into the complexities of Medicaid policy documents—specifically 1915(c) waivers. Traditionally, policy researchers have had to extract information from these documents manually, painstakingly combing through hundreds of pages of 𝐌𝐞𝐝𝐢𝐜𝐚𝐢𝐝 𝐝𝐚𝐭𝐚. This is often the most time-consuming part of the entire research process, acting as a massive bottleneck for policy analysis. What started as a support role quickly turned into a mission to solve the "PDF trap." These documents are the backbone of state healthcare safeguards, but because they are stored as static, semi-structured PDFs, cross-state analysis is incredibly difficult. My goal, to build a 𝐏𝐲𝐭𝐡𝐨𝐧-𝐛𝐚𝐬𝐞𝐝 𝐄𝐓𝐋 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞 to liberate this data into an analysis-ready Excel format, transforming weeks of manual labor into seconds of automated processing. 𝐓𝐡𝐞 𝐆𝐨𝐚𝐥: Automate the extraction of regulatory documents into a structured dataset to identify state-specific safeguards and transition plans. 𝐓𝐡𝐞 𝐓𝐡𝐢𝐧𝐤𝐢𝐧𝐠: I chose PyMuPDF (fitz) over simpler libraries because I needed to access page objects and text blocks directly. This was essential to maintain the context that a human researcher would normally catch during manual review. 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠𝐬: Data science in a research setting is often 90% "data unlocking." You have to bridge the gap between unstructured records and structured insights before any high-level modeling can begin. 𝐀𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧𝐬: This framework is a direct fit for any industry—legal, insurance, or compliance—where high-stakes data is currently buried in manual government filings. #DataScience #HealthIT #Python #ETL #PublicPolicy #Medicaid #Research #UTArlington

17

1 Comment

🎉 Just wrapped up Project 3 in my post-grad program in AI & Machine Learning for Business Applications at UT Austin! This time, we tackled a real-world classification challenge, predicting visa certification outcomes for foreign workers applying to the U.S. workforce. Think government-scale decisions, hundreds of thousands of applications, and an increasing demand for smarter, faster insights. ✅ What I built: I designed and tuned multiple machine learning models (Gradient Boosting, AdaBoost) using techniques like oversampling and hyperparameter tuning to optimize recall and F1 performance. The final model reached over 90% recall, critical in this use case where false negatives could mean missing out on qualifying candidates. 🔍 What I learned: From exploratory data analysis to ensemble learning and feature importance interpretation, this project sharpened my understanding of the full ML pipeline, especially how to balance interpretability with performance when decisions affect real people. ⚖️  Why it matters in pre-sales: At Eve, I help firms unlock the potential of AI tools purpose-built for plaintiff-side legal workflows. Projects like this one reinforce how important it is to not just “build AI,” but to tune it carefully to the unique shape of legal data and processes. Whether it's surfacing the right case details or predicting the likelihood of a favorable outcome, the ability to rigorously validate and explain model behavior is essential to delivering meaningful impact for our users. 📚 1 project closer to the finish line, and 1 step deeper into solving real problems with data. #MachineLearning #Presales #AIforBusiness #SolutionsEngineering #UTAustin #DataScience #LegalTech #AI #ML #PostGradJourney

19

1 Comment