class AnshikaMishra:

    name        = "Anshika Mishra"
    pronouns    = "she/her"
    location    = "Pune, Maharashtra"
    university  = "Savitribai Phule Pune University (SPPU)"
    degree      = "B.Sc. Data Science  |  CGPA: 8.61 / 10"
    email       = "anshikamishra.25000@gmail.com"

    experience  = [
        "📊 Data Science Intern @ Ernst & Young LLP",
        "🤖 AI Virtual Intern @ Infosys Springboard",
        "🔬 Project Contributor @ Dept. of Tech, SPPU",
    ]

    strengths   = [
        "End-to-end ML pipeline design",
        "Time-series forecasting (ARIMA, Prophet)",
        "NLP & LLM integration (FinBERT, Gemini, LLaMA)",
        "Multi-agent AI systems (LangGraph, ReAct)",
        "Reproducible, production-grade Python workflows",
    ]

    achievements = [
        "🏅 Elite Certificate — NPTEL Python for Data Science",
        "🎓 Selected — Infosys Springboard Pragati Cohort 3",
        "💎 SheFi Scholar — Cohort 1",
        "🏛️  Workshop @ IIT Bombay — AI for Professionals",
    ]

    motto = "Explore patterns → Build systems → Create impact"

FinBERT · XGBoost · MLflow · SHAP · Streamlit · SEC EDGAR

Scored 264 earnings transcripts across 40 S&P 500 tickers — built a 3-layer weighted NLP pipeline achieving 75% classification accuracy and a 2.17 Signal Sharpe Ratio.

• 🧠 3-layer pipeline: 40% hedging detection · 35% FinBERT sentiment · 25% vocabulary signals
• ⚡ XGBoost ensemble (WEIGHTS_V2) → +2.83pp accuracy improvement; validated across 10 ablation experiments
• 🔍 Identified hedging as critical signal with 18.5pp source-quality gap (Motley Fool vs SEC EDGAR)
• 📊 Streamlit dashboard with MLflow tracking + SHAP explainability + percentile-based signals

Python FinBERT XGBoost MLflow SHAP Streamlit yfinance SEC EDGAR

LangGraph · FastAPI · Groq/LLaMA · Pinecone · Redis · SSE

Production-grade 5-agent LangGraph pipeline: Task Planner → Code Generator → Tester → Debugger → Reviewer — autonomously plans, generates, tests, and reviews code end-to-end.

• ⚡ Real-time SSE streaming + Pinecone vector cache for code-pattern reuse
• 🔄 Redis pub/sub message queue with in-memory fallback + retry with exponential backoff (3 attempts)
• 🌐 Full FastAPI + Swagger UI REST API (5 endpoints incl. SSE stream + session management)
• 🛡️ Code review scoring: security · performance · maintainability

LangGraph FastAPI Groq LLaMA-70B Pinecone Redis SSE Python

Flask · LangGraph · SQLAlchemy · TF-IDF · Gmail API · Celery

Full-stack autonomous recruiting system with 4-agent LangGraph architecture — ingests candidates, scores, tiers, and sends Gmail-based recruiter loops via OAuth2 automation.

• 📊 Weighted NLP scoring: TF-IDF similarity + answer quality + GitHub scoring → FastTrack / Standard / Review / Reject tiers
• 🔁 Self-learning scoring loop that updates dynamic weights from hiring outcomes
• 🛡️ Anti-cheat layer: AI-answer detection + copy-ring clustering (TF-IDF cosine) + suspicious timing flags
• 🔧 Flask REST API (12 endpoints) + Celery/Redis async queue + full pytest coverage

Flask LangGraph SQLAlchemy TF-IDF Gmail API Celery Redis Groq

XGBoost · Random Forest · SHAP · Google Earth Engine · GeoPandas · K-Means

End-to-end geospatial ML platform predicting Urban Heat Island intensity using satellite + terrain data — deployable for any global city in 5–10 minutes.

• 🛰️ GEE data ingestion: MODIS/Landsat LST, NDVI, NDBI, ESA WorldCover, SRTM DEM
• 📍 K-Means UHI zone classification + Getis-Ord Gi* hotspot analysis + SHAP explainability
• 🗺️ CLI outputs: publication-quality infographics + interactive multi-layer HTML map
• 🎯 XGBoost predicts Land Surface Temperature with Heat Risk Scores (Low / Medium / High)

Python XGBoost Random Forest SHAP Google Earth Engine GeoPandas K-Means

ARIMA · Prophet · Random Forest · Scikit-learn · Streamlit · Plotly

Dual-model ML health platform: ARIMA + Prophet forecasting (2.3-day MAE) + Random Forest PCOS classifier (87.3% accuracy, 0.91 AUC-ROC).

• 🩺 15+ clinical features with cross-validation, stratified sampling, automated imputation
• 📊 5-module Streamlit platform: period logging · water tracking · risk assessment · analytics · nutrition insights
• 📈 Interactive Plotly dashboards + CSV export

ARIMA Prophet Random Forest Scikit-learn Streamlit Plotly

Pandas · NumPy · Scikit-learn · Pearson Correlation · Papermill

Born at EY — open-sourced for the community. Automated price segmentation across 157 retail segments × 3 channels — replaced 2+ weeks of manual Excel work with a sub-15-minute pipeline.

Python Pandas Pearson Clustering Papermill Excel Scikit-learn

 

"Exploring patterns in data and building intelligent systems that create real-world impact."

— Anshika Mishra

⭐ If any of my projects helped or inspired you, a star goes a long way! ⭐