Hi,

I'm Kamayani Rai

Data Scientist & AI/ML Engineer

Contact
Kamayani Rai

Work Experience

AI Data Engineer (Contract) Apr 2026 – Present

Handshake

  • Contributed human preference annotations to RLHF training pipelines by running blinded side-by-side evaluations of frontier LLM outputs across 10+ multi-stage agentic coding tasks (Docker, OpenCode); scored responses on correctness, agent behaviour, communication quality, and code style; preference scores fed directly into reward model training across Python, FastAPI, and SQL.
ML Research Engineer Dec 2024 – Mar 2026

Arizona State University

  • Managed end-to-end ML lifecycle across distributed YOLOv8 and Faster R-CNN training jobs on Sol via SLURM; implemented fault-tolerant checkpointing to handle mid-run interruptions on shared cluster resources.
  • Built a DICOM preprocessing pipeline (CLAHE normalization, study-level train/val splitting) with DVC versioning to track dataset snapshots against model checkpoints across training rounds.
  • Implemented a semi-supervised self-training loop using CVAT-annotated data; generated pseudo-labels from high-confidence predictions and retrained across two rounds, improving mAP from 0.54 to 0.82.
Software Engineer Mar 2022 – Sep 2023

DXC Technology

  • Built Python validation scripts on AWS Lambda to catch billing and tax logic errors across millions of DirecTV subscriber records before invoicing; reduced manual review effort by 50%.
  • Tested API performance under simulated high-concurrency load on DirecTV's subscriber platform; identified bottlenecks in request handling and data retrieval, reducing response latency by 35%.
  • Integrated automated schema and business rule validation into CI/CD pipelines; prevented revenue-impacting bugs from reaching production on every release.
  • Built Snowflake dashboards tracking SLA compliance, error rates, and billing anomalies; surfaced recurring failure patterns in weekly leadership reports, reducing defect resolution time by 15%.

My Projects

LLM Evaluation Platform

A mini RLHF evaluation platform built on GCP. The idea: instead of paying a commercial model to judge every LLM response, train your own reward model on human preference data and use it as the judge. Cheaper, faster, and you own it.

GitHub

Deal Sight AI: M&A Due Diligence Pipeline

Five-stage M&A due diligence pipeline using Claude API, FastAPI, and React; automated CIM analysis, EBITDA normalization, LOI generation, and agreement review — reduced a 4-hour analyst workflow to under 60 seconds per stage.

GitHub

ClinicalRAG: Medical Document Q&A Pipeline

RAG pipeline for clinical document Q&A using ChromaDB, sentence-transformer embeddings, and cross-encoder reranking; serves cited answers with faithfulness scoring and sub-100ms concurrent inference.

GitHub

OrbitShield: Geospatial Collision Risk Forecasting

Collision risk forecasting system on 845K+ satellite records using DBSCAN clustering, PostGIS, and an XGBoost/RF/LSTM ensemble; engineered orbital trajectory features for risk classification and time-series prediction.

GitHub

HonorHealth: Surgical Implant Inventory Optimization

Applied Holt-Winters and LSTM time-series forecasting across 13,000+ purchase orders to predict demand and identify $30K+ in excess inventory; built Tableau dashboards tracking procurement patterns across six hospitals.

GitHub

Real-Time Multi-Tenant Feature Platform

Continuously computes user-level features from streaming events and serves them with low latency for ML models, personalization engines, and analytics systems.

GitHub

ChipSafe: Ingredient Warnings via OCR & CV

AI-powered web app that processes uploaded food label images using OCR and computer vision, extracts ingredient lists, and alerts users to harmful or allergenic items in real-time.

GitHub

My Education

2025 · GPA 4.0

Arizona State University

M.S. Business Analytics (Big Data) — W. P. Carey School of Business

  • Machine Learning
  • Natural Language Processing
  • Business Process Analytics
  • Analytical Decision Modelling
2022 · GPA 3.68

Christ University

B.Tech, Electronics & Communication Engineering (Intelligent Systems)

  • OpenCV & Image Processing
  • Probability & Queuing Theory
  • Intelligent Systems specialization

My Skills

Programming & Frameworks

Python SQL (PostgreSQL) SSIS PySpark R OpenCV Pandas NumPy PyTorch Scikit-learn Pydantic

Machine Learning

Computer Vision YOLOv8 Faster R-CNN NLP Transformers Semi-Supervised Learning XGBoost Random Forest ARIMA Time-Series Forecasting Feature Engineering Distributed Training

ML Systems & Data Engineering

ETL Pipelines Apache Airflow Kafka Docker FastAPI AWS (S3, Lambda, SageMaker, Athena) Snowflake CI/CD SLURM DVC

LLMs & Analytics

RAG Pipelines Agentic Workflows Prompt Engineering ChromaDB Sentence Transformers Tableau Power BI CVAT

My Articles

Medium

I Built a Breast Cancer Detection System End-to-End

A technical walkthrough of building a complete breast cancer detection pipeline — from raw DICOM mammography data through YOLOv8 training, precision-recall tuning, and Grad-CAM visualization for clinical validation.

Read on Medium
Medium

M&A Due Diligence Takes Weeks. We Built Something That Does It in Minutes.

How we built DealSight AI at the Hanwha AI Center HACathon — a five-stage M&A due diligence pipeline using FastAPI, React, and the Claude API that compresses weeks of analyst work into minutes.

Read on Medium