Available for internships

San Diego, CA

Hello, I'm

Ishika Nahar

Applied AI · Data Science · Machine Learning

Vedu I love u. I'm a UC San Diego student working across applied machine learning, data science, and AI product development. I've used behavioral, healthcare, conversational, wearable, and biomedical data to train models, evaluate AI systems, and turn technical results into tools people can use. Currently at UC San Diego (B.S. Cognitive Science (Machine Learning & Neural Computation)), expected june 2027.

Try interactive demos All projects Open Résumé

Try before you scroll

Interactive demos on the homepage

Technical Skills

What I actually use

Grouped by the work I've done — applied ML, AI evaluation, analytics, and scientific data.

Applied Machine Learning

Building and evaluating models on real, messy data.

Supervised learning
Feature engineering
Model evaluation
Error analysis
Computer vision
Behavioral modeling
Experiment design

AI Systems & Evaluation

Making AI behavior inspectable, explainable, and trustworthy.

LLM tracing
AI evaluation
Prompt & output analysis
FastAPI integration
Structured knowledge systems
Human-in-the-loop validation

Data & Analytics

Turning data into decisions and clear recommendations.

Python
SQL
Pandas
Data visualization
Tableau
Power BI
Excel
Cohort & user analysis

Scientific & Healthcare Data

Working fluently with specialized, high-stakes data.

Behavioral & pose data
Biomedical imaging embeddings
Wearable time-series data
Healthcare interoperability formats
Biological & clinical domain interpretation

Featured Projects

Case studies with measurable depth

Problem → action → result. Tap Try demo to jump into the live homepage explorers.

InternshipJune 2026 – Present

Luveo Health

Luveo Health Compliance Copilot

Problem

Healthcare organizations must check clinical actions against interoperability standards and compliance rules. Luveo builds an AI copilot that reasons over these scenarios; rule execution, explanati…

Action

Mapped and extended architecture normalizing Epic, Pyxis, and Omnicell-style events against pharmacy rules.

Result

Eight synthetic scenarios return structured Pass, Warning, or Fail evaluations.

Tech stack

FastAPI
React
Langfuse
AI Evaluation

ResearchJune 2025 – Present

Sternson Lab, UC San Diego

Behavioral Machine Learning for Social Observation

Problem

Manually reviewing long behavioral recordings does not scale. The lab needed structured data from two interacting mice to study social observation and learning.

Action

Annotated 8,200 frames; corrected difficult failure cases.

Result

0.85 mean ROC-AUC with five-fold cross-validation.

Tech stack

Computer Vision
SLEAP
XGBoost
Behavioral Data

Owkin · K-Scope · 1st place

certify(prediction)

GROUNDED

Necessity

0.46

Sufficiency

0.51

Specificity

0.80

HackathonJuly 2026

Owkin Hackathon (K-Scope)

Biomedical Foundation Model Evaluation at the Owkin Hackathon

Problem

Biomedical foundation models produce strong predictions, but it is unclear whether those predictions depend on meaningful tissue representations or spurious signal.

Action

Built the local evidence-card dashboard UI so the team could demo certify(prediction) without a live GPU.

Result

1st Place, Owkin Rewiring Biology Hackathon.

Tech stack

Phikon-v2
H-optimus-0
Probes
Matched-random controls

View All Projects & Case Studies

Projects

Analyses I ran end to end

Results-first cards — open a case study for methods and limitations.

Model performance comparison across perturbations

Clean Macro F1

0.86

Independent2025

Robustness of Health Text Classification

TF-IDF vs LSTM vs DistilBERT under synonym, typo, and word-drop noise.

DistilBERTLSTMTF-IDF

Case study

GitHub

ROC-AUC

0.884

Independent2025

Neural Decoder for Behavioral Choice

Decode left vs right choice from Neuropixels population activity.

PCALogistic RegressionPermutation testing

Case study

GitHub

Ideas worth exploring

Prototypes still in motion

Live UI previews — open the homepage demos or the case study.

Live UI

IndependentIdea2025 – Present

MOMentum

Context-aware personal AI — RAG + structured user state for proactive nudges.

Case study

Live UI

IndependentIdea2026

Style-Adaptive Extraction

Interactive exploration of pathology-style features and adaptive extraction paths.

Case study

Professional Experience

Research, internships, and builds

Roles where I shipped models, evaluations, and product-facing AI systems — with case studies linked where they exist.

Machine Learning Intern

Luveo Health

June 2026 – Present

Extending an AI compliance copilot that turns structured healthcare-compliance results into clear, actionable explanations.

Key achievements

Mapped and extended compounding support across domain models, field semantics, API contracts, and tests.
Connected a React clinical sandbox to an authenticated FastAPI backend across HL7 v2.5 / FHIR R4 scenarios.
Instrumented the LLM explanation layer with Langfuse for end-to-end tracing of prompts, tools, and failures.

Open case study

Data Engineer Research Assistant

UC San Diego Neurobiology Lab (Sternson Lab)

June 2025 – Present

Human-in-the-loop computer-vision pipeline to quantify social observational learning in mice.

Key achievements

Trained SLEAP multi-animal pose models on 8,200 annotated frames across three labeling batches.
Engineered trajectory features, aligned them with ELAN annotations, and trained XGBoost (0.85 mean ROC-AUC, 5-fold CV).
Built raster visualizations and a PyQt review GUI; contributed analyses presented at SfN 2025.

Open case study

Hackathon Teammate — Biomedical Model Evaluation

Owkin Hackathon (K-Scope)

July 2026

1st-place team building a certify(prediction) workflow for pathology foundation-model evidence cards.

Key achievements

Built the evidence-card dashboard UI for demoing probe / necessity / specificity results without a live GPU.
Collaborated on framing evaluation beyond accuracy with matched-random controls and structured caveats.

Open case study

Product & Content Strategy Intern

Skillsoft Corporation

July 2024 – September 2024

Analyzed CAISY beta conversation and learner-engagement data to inform prompting and audience targeting.

Key achievements

Turned conversation logs into product recommendations for prompts and targeting.
Built Tableau / Power BI views that made cohort patterns usable for the product team.

Open case study

View Complete Experience Timeline

Let's build something reliable together

Open to Applied AI, ML, and Data Science internships for Fall 2026 and Summer 2027 — especially teams that care about evaluation, not just demos.

Get In Touch Download Résumé

inahar@ucsd.edu