JM Résumé
Johaan Mannanal
Menu

I love building.

... , Muay Thai, performing magic, and chocolate lava cake 😋

View work Résumé

Experience

Product engineering, applied machine learning, and research and consulting at Purdue.

  1. Aug. 2021 to present

    Founder and Product/Engineering Lead, Midnight

    I led the rebuild and coordinated 25 contributors. My engineering work spans course ingestion, grounded tutoring, visual lessons, comprehension checks, and multi-provider AI routing. I bring product planning, implementation, and student feedback into the same development cycle.

  2. Sep. 2026 to present

    AI Stroke Detection Researcher & Consulting Project Member, VITAL at Purdue

    I explore AI-based stroke detection on a five-person research team. Separately, I support Vantro Research as a member of a six-person consulting project team.

  3. Jul. 2022 to Aug. 2024

    Machine Learning Intern, Project Read

    I helped develop AI-assisted phonics and decoding practice for early readers. On a five-person technical team, I built and evaluated models that map generated speech sounds to written letters, refining the algorithm logic and test cases.

More projects

Health research, learning tools, and telemetry.

More projects, contributions, and links
Project Role & context What it does Status Links
03 Wearable Health ML Research Owned preprocessing, model comparisons, cross-validation, and reproducibility. Research with Jevelson Simenthy, a Penn State IT instructor; later iOS work with Yash Piratla. Explored Apple Watch and synthetic signals to study longer-term heart-health risk patterns in older adults. The public pipeline rebuild uses synthetic data only, with held-out evaluation and checks for data leakage. 0.99 macro-F1 on the synthetic four-class benchmark. The generated classes are separable by design; this is not evidence of clinical or real-patient performance. Open Source
Source and case studyModel card & limits
04 Snap-a-Mistake Solo developer · Expo/React Native and Fastify Reads handwritten math, checks the transcription against the image, and verifies the proposed mistake When the checks disagree, the app flags the result as uncertain instead of presenting it as a confirmed mistake. Open source
Source
05 Prometheus Built the React/Convex app, source-grounded card generation, learner-approved adaptations, and FSRS review flow for the IncludAI Neurodiversity Hackathon. Creates flashcards from course material, lets learners edit and approve them, and schedules reviews with FSRS Learners can adjust how a card is presented while keeping its learning target. Live prototype
Live demoSource
06 Motorsport Telemetry Analytics Solo developer · Python analysis through deployed frontend Static telemetry web app using real FastF1 data and a pace model evaluated on held-out stints 0.40s MAE on held-out Monza stints · 46% below a mean-prediction baseline Live
Live demoSource
03 Wearable Health ML Research Exploring wearable signals and longer-term heart-health risk Open Source
My role
Owned preprocessing, model comparisons, cross-validation, and reproducibility. Research with Jevelson Simenthy, a Penn State IT instructor; later iOS work with Yash Piratla.
What it does
Explored Apple Watch and synthetic signals to study longer-term heart-health risk patterns in older adults. The public pipeline rebuild uses synthetic data only, with held-out evaluation and checks for data leakage.
Details
0.99 macro-F1 on the synthetic four-class benchmark. The generated classes are separable by design; this is not evidence of clinical or real-patient performance.
Source and case studyModel card & limits
04 Snap-a-Mistake Helps identify the first unsupported step in handwritten math Open source
My role
Solo developer · Expo/React Native and Fastify
What it does
Reads handwritten math, checks the transcription against the image, and verifies the proposed mistake
Details
When the checks disagree, the app flags the result as uncertain instead of presenting it as a confirmed mistake.
Source
05 Prometheus Course-grounded flashcards that learners control Live prototype
My role
Built the React/Convex app, source-grounded card generation, learner-approved adaptations, and FSRS review flow for the IncludAI Neurodiversity Hackathon.
What it does
Creates flashcards from course material, lets learners edit and approve them, and schedules reviews with FSRS
Details
Learners can adjust how a card is presented while keeping its learning target.
Live demoSource
06 Motorsport Telemetry Analytics Monza pace model: 0.40s MAE on held-out stints Live
My role
Solo developer · Python analysis through deployed frontend
What it does
Static telemetry web app using real FastF1 data and a pace model evaluated on held-out stints
Details
0.40s MAE on held-out Monza stints · 46% below a mean-prediction baseline
Live demoSource

About

Builder and data scientist

I’m a Data Science student at Purdue and the Founder of Midnight, where I build tutoring tools around students’ own course material. Through VITAL, I also work on AI stroke-detection research and a consulting project supporting Vantro Research.

Education
B.S. in Data Science, Purdue University · Expected May 2029
Focus
Applied machine learning, health research, and product engineering
Analytics
Python, SQL, Power BI, and Microsoft Excel
Based in
Seattle, Washington · Studying at Purdue in West Lafayette, Indiana
Outside work
Muay Thai, piano (classical and jazz), magic tricks, and chocolate lava cake

Contact

Start a conversation

If you’re hiring an intern, mentoring an ambitious builder, or assembling a research team, I’d like to hear what you’re working on.

jmannana@purdue.edu