CSE (AI & ML) · PCCOE Pune · Third Year

Ananya
Rajankar

I build intelligent systems — from predictive ML pipelines to AI-powered products. Two published papers, a MeitY-sponsored research internship, and a habit of shipping things end-to-end.

View Work Get in Touch

Published Papers

91.4%

Best Model Accuracy

8.85

Last Sem GPA

01 /

About

I'm a third-year CSE (AI & ML) student at PCCOE, Pune — working at the intersection of research and real product-building. I don't just study AI; I deploy it.

My background spans the full stack: data pipelines, ML/DL models, NLP systems, full-stack web development, and research-level contributions to forensics, agriculture, and education tech.

I've been lucky to have a MeitY-sponsored research internship at IIIT Pune, two peer-reviewed publications, and multiple shipped products under my belt — all before graduating.

Outside the technical work, I think like a product person: I spot friction, define the problem, and build the fix. NomNom is a recent example — a food-matching app for groups, built solo because the problem was genuinely annoying.

Currently B.Tech CSE (AI & ML), Year 3

Institution PCCOE, Pune

CGPA 7.91 overall · 8.85 last sem

GitHub github.com/Rajankarananya

Phone +91 9604416699

Location Pune, Maharashtra

02 /

Skills

Languages

PythonRSQL C++JavaScript

ML / Data Science

Scikit-learnXGBoostSHAP PandasNumPyEDA Feature Eng.ROC-AUC

Deep Learning & NLP

PyTorchTensorFlowKeras BERTLSTMRAG LangChainOllamaOpenCV

Web & Backend

React.jsNode.jsFastAPI Express.jsREST APIsD3.js

Databases

PostgreSQLMySQLMongoDB PostGISPL/SQL

Tools & Other

GitFigmaPostman JupyterUnityBlender ARCore

03 /

Projects

✦ Featured · Product

NomNom

Hinge, but for choosing what to eat

Solving the oldest group problem: nobody can agree on food. NomNom is a swipe-based food-matching app for friends, couples, and groups — users swipe on food options individually and the app reveals what everyone actually agrees on. Built and shipped solo, end-to-end.

React.jsNode.jsVercel

Live Site →

Problem → Solution

"What do you want to eat?"
"I don't know, whatever you want."

↓

Everyone swipes. App surfaces the match. Decision made.

AI · EdTech

Skillo

Predictive AI for Invisible Learning Gaps

End-to-end predictive analytics on 2,290 student quiz samples. XGBoost classifier with BERT-based semantic features, RAG-powered retrieval, and real-time educator tooling.

91.4%

Accuracy

0.963

ROC-AUC

0.923

Top-3 Recall

XGBoostBERTRAGFastAPIReact

AI · AgriTech

Annadata Connect

AI-Powered Crop Insurance & Fraud Detection

Fine-tuned EfficientNet-B0 for crop damage classification. PostGIS-based fraud detection by cross-referencing GPS metadata against satellite damage maps. Multilingual RAG chatbot for farmers. District-level D3.js risk heatmaps.

EfficientNet-B0PostGISLangChainD3.js

HealthTech

Alzheimer's Support App

Full-stack care coordination platform

Full-stack web application for Alzheimer's patients and caregivers. JWT auth, caregiver coordination module, cognitively friendly UI designed around UX research on cognitive impairment.

Node.jsExpress.jsMySQLJWT

04 /

Experience

Nov 2025 – Mar 2026

IIIT Pune
MeitY-Sponsored

Research Intern

Developed ML/DL models for automated forensic analysis and artifact recovery from storage drives using pattern recognition on raw disk data.
Built GUI interfaces to visualise forensic outputs; performed LevelDB storage file analysis to extract and interpret structured digital artifacts from unstructured disk images.
Contributed to next-generation data carving techniques using deep learning for extracting structured information from unstructured storage media.

Jun 2025 – Aug 2025

EventicHub Pvt. Ltd.

Frontend Development Intern

Built responsive, scalable web interfaces with HTML, CSS, JavaScript, and Django templates; focused on usability and cross-platform compatibility.
Integrated frontend modules with backend APIs and databases for dynamic application behaviour.
UI/UX optimisation via prototyping and Figma-based design iteration in a Git-based workflow.

05 /

Research

📄

Published · IJSREM

YOLOv8-Based Real-Time Object Detection for Transport Flow Optimisation

Trained a YOLOv8 model on a custom annotated transport dataset with full ML pipeline: data collection, augmentation, preprocessing, training, and benchmarking. Real-time inference pipeline using OpenCV for live video stream processing, with measurable improvements in vehicle detection accuracy and traffic flow analysis.

YOLOv8PyTorchOpenCV

⚡

Published · IEEE ICCUBEA 2025

Enhanced Battery Management System (BMS) for Formula SAE Electric Vehicles

Designed an advanced BMS using STM32 master-slave microcontroller architecture for distributed real-time cell monitoring across high-voltage battery packs. Implemented fault detection for overvoltage, undervoltage, overcurrent, and thermal runaway conditions with CAN Bus and isoSPI protocols.

STM32Embedded C/C++CAN BusisoSPI

🩺

Project

Hybrid Ensemble Model for Early-Stage Diabetes Prediction

Hybrid ML model combining Logistic Regression, Random Forest, SVM, and Decision Trees via soft-voting ensemble. Thorough EDA, stratified k-fold cross-validation, and Bayesian hyperparameter tuning. Evaluated across ROC-AUC, F1, precision-recall, and confusion matrices.

Scikit-learnPandasEnsemble Methods

06 /