abhay's little corner✳
A bit about meLet's connect
A HEY, I'M ABHAY WELCOME TO MY PINBOARD

A little bit of code.
A lot of curiosity.

Software, data, and the things in between. A collection of what I build, learn, and notice.

Building. Learning. Looking around.My experience Take a look around
THE WORK & THE LEARNING

Experience, with perspective.

Education

I like the point where a hard technical problem becomes something another person can actually use. My work has moved between medical imaging, machine learning, data platforms, and teaching, but the thread is the same: make the data trustworthy, test the result, and build a useful way to act on it.

EXPERIENCE2021 — PRESENT

  1. University of Maryland, College Park (opens a new tab)

    Current

    Teaching Assistant, Principles of Data Science (602)

    College Park, Maryland

    Helping students ask clear questions, check their data, and explain what the results mean.

    Read the storyA little less — Teaching Assistant, Principles of Data Science (602) at University of Maryland, College Park

    I currently support students in Principles of Data Science. Teaching keeps me close to the habits I rely on in my own work: ask a clear question, check the data, and explain what the result does and does not show.

    Data scienceTeaching

  2. Longeviti Neuro Solutions (opens a new tab)

    Biomedical Software Engineer Intern

    Baltimore, Maryland

    An MRI-to-ultrasound workbench that made model outputs inspectable, with patient-level evaluation.

    85,000+ imaging files · 114 patients
    Read the storyA little less — Biomedical Software Engineer Intern at Longeviti Neuro Solutions

    I helped turn an MRI-to-ultrasound research pipeline into a workbench where the team could inspect results instead of taking a model’s output on faith. I built the Python and FastAPI backend, connected it to four synchronized image views, and made each case’s outputs available for review. I also organized more than 85,000 imaging files from 114 patients and evaluated model performance using patient-level holdouts, so results reflected new cases rather than repeated images from the same patients.

    PythonFastAPIMedical imaging

  3. Cornell University / Break Through Tech AI (opens a new tab)

    Lab Facilitator and Teaching Assistant

    Remote

    Coaching students from “the code runs” to “I understand what the result means.”

    100+ students · 9 hands-on labs
    Read the storyA little less — Lab Facilitator and Teaching Assistant at Cornell University / Break Through Tech AI

    I coached more than 100 students through Python, SQL, and machine learning across nine hands-on labs. My favorite part was helping someone move from “the code runs” to “I understand what the result means.” I also helped design an eight-week, Git-based curriculum with code review and practical feedback; assessment scores improved by 18%.

    PythonSQLMachine learning

  4. Accenture (opens a new tab)

    AI/ML Analyst

    Hyderabad, India

    Production anomaly detection, controlled model evaluation, and faster document search for claim reviewers.

    10M+ records/day · 92% precision
    Read the storyA little less — AI/ML Analyst at Accenture

    I worked on models and tools that had to hold up in everyday operations, not just in a demo. That included anomaly detection over more than 10 million records a day, with 92% precision and 30% less false-positive review work. I designed A/B tests against production baselines, analyzed lift and segment results, and checked retrained models in shadow mode before promotion. I also piloted a document-search tool with claim reviewers against manual search; average policy lookup time fell 40%, and claim resolution became 20% faster.

    Machine learningA/B testingDocument search

    Data Engineering Associate

    Hyderabad, India

    Reliable data foundations for reporting and machine learning, with careful validation before cutover.

    2M+ records/day · 99.9% reliability
    Read the storyA little less — Data Engineering Associate at Accenture

    I built the data foundation behind reporting and machine learning: Python, SQL, and Spark pipelines handling more than 2 million records a day at 99.9% reliability. I helped bring more than 12 sources into a shared Delta Lake layer and checked its outputs against the legacy system before cutover. One of my most useful fixes was small but consequential: I found a SQL join that overstated a daily metric by 8%, explained the 14 days it affected, and corrected and backfilled the history within three hours.

    PythonSQLSpark

  5. Rajiv Gandhi Proudyogiki Vishwavidyalaya (opens a new tab)

    Research Assistant

    Bhopal, India

    Fine-tuning language models and documenting what changed across controlled comparisons.

    15+ controlled comparisons
    Read the storyA little less — Research Assistant at Rajiv Gandhi Proudyogiki Vishwavidyalaya

    I explored how model choices change real results, fine-tuning language models on more than 1 million text samples and running over 15 controlled comparisons. The work improved classification accuracy by 15–20% over the baseline and taught me to document what changed, how it was measured, and where a result might not generalize.

    NLPResearchModel evaluation

  6. Stellar Search (opens a new tab)

    Data Science Intern

    Remote

    Turning predictive analysis into Tableau views and recommendations people could act on.

    50,000+ client records
    Read the storyA little less — Data Science Intern at Stellar Search

    I compared regression and classification approaches on more than 50,000 client records, improving prediction accuracy by 12%. I then turned the findings into Tableau views and plain-language recommendations for stakeholders, helping them make decisions faster.

    Data scienceTableauPredictive modeling

Across these roles, I have learned to be as interested in the question and the measurement as I am in the model or code. Good work makes the next decision clearer for the person using it.

A bit of everything8

Projects, thoughts, and the occasional pretty view.

All pins▦ Boards
A WORK IN PROGRESS, JUST LIKE ME
Featured
PROJECT

xUMD: Experience UMD

A campus companion built for Terps: discover events on a live map, connect with students, and keep campus life in one place. Coming soon to iOS, Android, and the web.

Mobile AppsCampus LifeProduct Engineering
There's always something new around the corner.Back to the top ↑