ReInsureSight
A geospatial flood loss estimation engine using LightGBM and Google Earth Engine across 1,202 NatCat events in 131 countries, with MAPIE conformal prediction achieving 87.1% empirical coverage on held-out test data.
Hi, I'm Aishrica
Learner.
Builder.Data Scientist.
I'm an M.Sc. Data Science candidate at IIIT Lucknow, and I build production ML systems — from an LSTM heatwave-forecasting pipeline for the Uttar Pradesh government to RAG-based learning tools. This is my work, one project at a time.
Student → Data Scientist
M.Sc. Data ScienceIIIT Lucknow
CRO, UP Govt.Data Scientist Intern
Top 5%Kaggle Datasets Expert
150+LeetCode DSA & SQL solved
My journey
How I got here.
From an integrated B.Ed. classroom to government climate models — every step looked unrelated to the last, until I could see the thread running through all of it: turning raw data and complex systems into something people can actually use.
01
2021 – 2025
Foundations
I did my B.Sc. B.Ed. (Integrated) at DAV University, Jalandhar (GPA 8.35) — building the grounding in math, statistics, and how people actually learn that later shaped how I approach modeling and teaching alike.
02
2025 – 2027
The Pivot to Data Science
I began my M.Sc. in Data Science at IIIT Lucknow (GPA 8.45), specializing in predictive modeling, NLP, and time-series forecasting — trading the classroom for end-to-end ML pipelines.
03
Sep 2025 – Jun 2026
Climate Resilient Observatory
As a Data Scientist Intern, I built a multivariate LSTM pipeline for heatwave forecasting on live UP government climate data, achieving 0.82 recall on extreme heat event detection, with XGBoost residual correction and SQL-based data quality validation.
04
Jan 2026 – May 2026
Teaching Assistant
I conducted PyTorch labs on CNNs, RNNs, and backpropagation, and delivered BERT fine-tuning sessions for sentiment analysis and text classification — supporting 10+ students' NLP projects end-to-end at IIIT Lucknow.
05
2025 – Present
Shipping in Public
ReInsureSight, SmartPlate, Saarthi-AI, and an AI Learning Assistant — I ship real, working projects on GitHub and Hugging Face, public and checkable, not just listed on a resume.
What I work with
My technical toolkit.
Not a buzzword list — this is the actual stack behind my CRO heatwave forecasting pipeline, ReInsureSight, SmartPlate, and the RAG systems below.
Languages & Libraries
Experimentation & Statistics
Machine Learning
Deep Learning
RAG & LLM Tooling
Visualization
Databases & Tools
Right now
What I'm building.
Since September 2025, I've been building climate forecasting models and teaching the next batch of ML students at IIIT Lucknow — here's the actual work behind that.
01
Climate Time-Series Forecasting
A multivariate LSTM pipeline for heatwave forecasting on live CRO climate monitoring data across Uttar Pradesh — 0.82 recall on extreme heat event detection.
02
Residual Correction & Data Quality
XGBoost-based residual correction layered on LSTM forecasts, validated against real-time government data, plus automated SQL data-quality checks on historical climate observations.
03
RAG & Agentic Systems
Retrieval-augmented learning assistants and commute-planning agents — LangChain, ChromaDB, vector search, and LLM orchestration built end-to-end and shipped on GitHub and Hugging Face.
04
Teaching & Mentorship
PyTorch labs on CNNs, RNNs, and backpropagation, plus BERT fine-tuning sessions for sentiment analysis and text classification — supporting 10+ students' NLP projects end-to-end.
11Public GitHub repos
10GitHub followers
2Hugging Face Spaces
3Certifications & credentials
Things I've built
From idea to shipped code.
Public, checkable projects — not just claimed on a page, but live on GitHub and Hugging Face.
SmartPlate
A fine-tuned Vision Transformer (ViT-Base) over 4,000 images across 80 Indian food categories (75.19% test accuracy), wired to a FastAPI inference pipeline returning serving-size-adjusted calorie estimates.
AI Learning Assistant
A retrieval-augmented learning assistant for ML/DL concepts — LangChain, ChromaDB, Groq Llama-3.3-70B, and HuggingFace embeddings, with source-cited, page-referenced answers instead of generic responses.
Saarthi-AI
A predictive commute-planning agent for Lucknow that fuses traffic, weather, festival, and event signals into a deterministic risk score, then narrates optimal departure windows via an LLM.
Cyberbullying Classifier
A multi-category (6-class) cyberbullying detection system for social media tweets — TF-IDF, spaCy preprocessing, and a Naive Bayes / Logistic Regression / Random Forest / XGBoost comparison with hyperparameter tuning.
LeetCode Solutions
A running archive of solved Data Structures, Algorithms, and SQL problems — the working record behind the 150+ problems solved on LeetCode.
View on GitHub →Deep dives
The story behind the code.
Six full write-ups — what I researched, what already existed, why it wasn't enough, and exactly how I built mine instead. Not a changelog, the actual reasoning.
ReInsureSight
Why most catastrophe models are confidently wrong in the countries that need them most — and how I built one that admits its own uncertainty instead.
Read the case study →Forecasting Heatwaves at CRO
Why "95% accurate" is the wrong goal for an early-warning system, and how an LSTM + XGBoost pipeline got tuned to actually catch the days that matter.
Read the case study →SmartPlate
Why mainstream calorie apps can't tell dal from sambar, and how a fine-tuned Vision Transformer fixed that for 80 Indian food categories.
Read the case study →AI Learning Assistant
What happens when a TA gets tired of answering the same question a hundred times, and builds a RAG system that cites its sources instead of guessing.
Read the case study →Saarthi-AI
Why deterministic math should come before the LLM in an agent, not after — and how that design choice shaped a commute-planning agent for Lucknow.
Read the case study →Cyberbullying Classifier
Why "is this bullying, yes or no" is the wrong question for real moderation, and how a 6-class model answers the one that actually matters.
Read the case study →Credentials
Proof, not just a portfolio.
Formal credentials, algorithmic rigor, and applied ML experience behind the data science work above.
Algorithms
LeetCode
150+ Data Structures, Algorithms, and SQL problems solved, strengthening the algorithmic problem-solving behind production pipeline design.
View Profile →Credentials
Kaggle & Certifications
- Kaggle Expert in Datasets — Top 5%
- SQL (Advance) Certificate — HackerRank
Education
IIIT Lucknow
M.Sc. Data Science, 2025 – 2027 (in progress), GPA 8.45. B.Sc. B.Ed. (Integrated), DAV University, Jalandhar, 2021 – 2025, GPA 8.35.