Hi, I'm Payal

Designing AI systems with engineering discipline.

About

Final year at IIT Roorkee building production AI/ML systems — RAG pipelines, containerized inference services, and semantic retrieval. Open to ML/AI Engineering roles.

Latest Writing

Jan 28, 2026

Linear Algebra for Machine Learning

Read on Substack ↗
Jan 14, 2026

Entropy vs Gini

Read on Substack ↗
Jan 07, 2026

Teaching Models to Behave Like Humans

Read on Substack ↗

View all articles

payalkanyan.substack.com →

Work Experience

Center For Sustainable Energy, IIT Roorkee
Research Intern
May 2025 – July 2025 ▸
  • Built predictive models for ice adhesion analysis.
  • Achieved 92% prediction accuracy using Random Forest & XGBoost.
    • GlueX
    Software Intern
    Jan 2025 – March 2025 ▸
    • Developed Python integrations for 10+ APIs.
    • Achieved 95.5% test accuracy with automated unit testing pipelines.

    Skills

    Core ML
    Machine Learning
    Deep Learning
    CNNs
    Transformers
    LLMs
    Frameworks & Libraries
    PyTorch
    LangChain
    LangGraph
    FastAPI
    Streamlit
    AI Systems & Architectures
    Agentic AI
    Multi-agent Workflows
    RAGs
    Engineering
    Python
    SQL
    Docker
    AWS

    Check out my latest work

    I've worked on a variety of AI/ML projects, from Classification to RAG systems. Here are a few of my favorites.

    Personal Knowledge OS

    Stateful LangGraph agent enabling autonomous tool-routing between local databases and web search. Features Qdrant vector search, Neo4j graph traversal, and a Chrome extension for web context capture.

    Agentic AILangGraphQdrantNeo4jCeleryPostgreSQL
    Source

    ScholarLens

    Semantic retrieval system using BERT embeddings and MLflow. 27% higher accuracy, 35% lower latency.

    BERTMLFLowSemanticSearch FAISS
    Source

    Kora

    RAG-based AWS S3 assistant using HuggingFace and ChromaDb.

    RAG Systems FastAPIChromaDB
    Source

    Roster Email Parser

    Hybrid OCR + NER pipeline for healthcare emails. 86% extraction accuracy, 60% higher throughput.

    OCRNER NLPPython
    Source

    Air travel Delay Prediction

    Explored delay patterns and root causes via EDA. Build & compare regression models to predict (a) whether a flight will be delayed and (b) its expected delay in minutes

    NumPySci-kit LearnXgBoost LigthGBM
    Source

    Credit Card Behaviour Prediction

    Analysed anonymized historical data of over 30,000+ customers to build an interpretable and high‑performance credit risk model.

    EDAFeature EngineeringHyperparameter Tuning Evaluation Metrics
    Source

    Education

    Indian Institute of Technology, Roorkee
    Bachelor of Technology in Chemical Engineering
    2022 – 2026

    Get in Touch

    Want a quick chat? DM me on X