PORTFOLIO · HYDERABAD, IN
AI/ML Engineer
THE ENGINEER BEHIND THE PIPELINES
I'm Kevin Costner — an AI/ML engineer in Hyderabad building agentic RAG, healthcare ML, and full-stack systems that ship. GitHub: kevincostner17.
Hands-on with Cursor and Claude for AI-assisted development. Open-source contributor to FreshCode-Org/freshdata (and my fork kevincostner17/freshdata). Proven in healthcare classification and explainable cleaning pipelines.
Every build starts with the same question: what does the data actually support — and how do we ship it with an audit trail?
Autonomous agents that decide when to retrieve, then ground answers in document context.
Classification pipelines trained, evaluated, and leakage-checked until the metrics hold.
Streamlit demos, Next.js apps, and CI/CD so prototypes leave the notebook.
THE TOOLBOX
WHERE THE WORK COMES FROM
LLMs, Generative AI, and production-scale model deployment. Built classification and RAG-oriented systems with Python, Scikit-learn, and modern LLM tooling.
Explainable Python data-cleaning library for pandas: missing-value handling, outlier detection, dtype repair, and duplicate resolution with a clear audit trail. Hardened Core API outlier detection for boolean columns on mixed-type DataFrames.
Glaucoma Detection System — predictive healthcare pipeline on 5,000+ patient records. Benchmarked SVM, KNN, Random Forests, and Decision Trees; automated cleaning and precision-recall evaluation across demographics.
SELECTED WORK
PyPDF extraction, chunking, MiniLM embeddings in ChromaDB, and a CrewAI retrieval tool.
Assembled an agentic document-analysis system with 800-char / 150-overlap chunking. Exposed top-3 semantic retrieval so an Ollama Llama 3.2 agent autonomously decides when to search and grounds financial answers in retrieved context.
30-day diabetic readmission risk ranking on UCI inpatient data with a live Streamlit demo.
Capacity-constrained outreach scenarios with isotonic calibration and subgroup fairness audits. Gradient-based attributions, model card, and a deployed Streamlit demo.
Explainable semantic cleaning for pandas — automated repairs, local AI intelligence, and a full audit trail.
Contributed Core API outlier-detection fixes for boolean columns, reviewed pipeline edge cases on real datasets, and shipped clean Python/Pandas PRs. Upstream: FreshCode-Org/freshdata.
Tech jobs marketplace for seekers and employers — search, filters, posting flow, REST API, Vercel CI/CD.
NEXT.JS · TYPESCRIPT · TAILWINDPharmacy management MVP with FastAPI services, React POS, RBAC, and AI anomaly scoring.
FASTAPI · POSTGRESQL · REDISMulti-label healthcare disease risk ANN with TensorFlow and a Streamlit demo.
TENSORFLOW · STREAMLITProduct catalog site for Dhanalakshmi Security Service CCTV offerings.
HTML · CSS · LIVE SITEThis site — white/navy cinematic scroll portfolio with HQ video backdrops, deployed on Vercel.
HTML · GSAP · VERCELPython ML assignment work covering classification workflows and evaluation hygiene.
PYTHON · SCIKIT-LEARNPROOF OF WORK
1st among 140 teams; optimized algorithmic logic in a 24-hour sprint.
IIT Madras (NPTEL) — November 2023 – April 2024. Neural networks and deep learning foundations.
IIT Kharagpur (NPTEL) — June 2024 – November 2024. IoT systems and connected applications.
B.Tech in Artificial Intelligence & Machine Learning
CGPA 8.12/10.0 · August 2021 – May 2025 · Hyderabad, Telangana · Coursework: Neural Networks, Reinforcement Learning, Computer Vision, NLP, Data Mining, DBMS, OS
Data Science & AI Program
LLMs, Generative AI, and production-scale model deployment
Deep Learning
Neural networks and deep learning
Internet of Things
IoT systems and applications
Ready to ship the next pipeline? Whether it's agentic RAG, healthcare ML, or a production demo — let's talk.