Abhijith S Kurup

About

AI/ML Engineer and fresh B.Tech CSE (AI & ML) graduate with a production-grade portfolio spanning LLM finetuning, RAG systems, healthcare AI, computer vision, generative AI, and predictive analytics — all built from scratch and deployed. Fine-tuned Mistral-7B using QLoRA (98.9% token accuracy, published on Hugging Face); built a full RAG chatbot using LangChain, ChromaDB, and LLaMA 3.1 8B; a clinical symptom classifier; a medical document summarizer; a published CNN biometric authentication system (IJARCCE, Impact Factor 8.471); a live government forecasting dashboard (98.2% accuracy); and a generative AI image synthesis pipeline. Strong Python programmer with hands-on experience across the complete AI/ML stack — from data engineering and model training through LLM fine-tuning, orchestration, vector database integration, and production deployment. District-level Government of Kerala AI Quiz winner. Proven ability to build, ship, and document enterprise-grade AI solutions independently.

Experience

• Machine Learning Intern
  Unified Mentor Pvt. Ltd.
  Apr 2026 – Jun 2026

  Built and deployed an end-to-end predictive analytics system on 1,075 days of real U.S. government operational data (HHS Unaccompanied Alien Children program) — achieving 98.2% forecast accuracy (MAPE 1.84%) on unseen data, measurably outperforming 5 baseline models. • Engineered 9 statistical features from scratch (temporal lag variables, rolling aggregates, pressure indicators, surge flags); applied strict time-based train-test splitting; benchmarked 6 models using MAE, RMSE, MAPE, and 30-day walk-forward cross-validation. • Deployed a 6-page live production dashboard on Streamlit Cloud with 90-day forward forecasting, confidence intervals, surge early warning (7+ day advance notice), and capacity breach probability scoring; maintained full version-controlled codebase on GitHub.

• AI/ML Intern
  InnoKnowvex (MSME & StartupIndia Verified)
  Nov 2025 – Dec 2025

  Designed and trained a Generative AI image synthesis pipeline from scratch — applied prompt engineering to optimise generative model outputs, evaluated image quality across training iterations, and delivered clean documented Python code as the primary internship deliverable. • Completed InnoKnowvex AI Training Programme (Sep–Nov 2025) before the internship, demonstrating proactive continuous learning

• AI & Robotics Intern
  Srai Smart Solutions Pvt. Ltd. (AccelMove)
  May 2023 – Jun 2023

  Completed two consecutive 15-day internships in Robotics & AI at Technopark's EdTech Product Development Unit; collaborated with cross-functional engineering teams on AI product development and attended a one-day AI Workshop.

Education

• B.Tech / B.E.
  College of Engineering Kottarakkara · APJ Abdul Kalam Technological University
  Computer Science and Engineering with Specialization in Artificial Intelligence and Machine Learning · Oct 2022 – Apr 2026
• 12th / Higher Secondary (Class 12)
  KENDRIYA VIDYALAYA AMC LUCKNOW
  Physics, Chemistry, Mathematics, Biology · Mar 2020 – Apr 2021
• 10th / Secondary (Class 10)
  KENDRIYA VIDYALAYA AMC LUCKNOW
  CBSE 10th · Mar 2018 – Apr 2019

Skills

• Artificial intelligence
• Computer Vision
• Deep Learning
• FastAPI
• Git
• APIs
• Python
• Machine Learning
• Data Science
• Java
• REST API
• Artificial Intelligence Markup Language (AIML)
• PyTorch
• TensorFlow
• Docker
• Keras
• LangChain
• Software Development
• CI/CD
• NLP
• RAG
• LLM
• ML Ops
• Hugging Face Transformers
• Chromadb

Projects

• Mistral-7B Medical Instruct — QLoRA Fine-tuning ↗
  Mistral-7B, QLoRA, PEFT, Hugging Face, PyTorch, Python

  Fine-tuned Mistral-7B-Instruct using QLoRA (4-bit NF4 quantization) on a medical Q&A dataset — achieved 98.9% token accuracy. Published model on Hugging Face Hub.

• RAG Chatbot — LangChain + LLaMA 3.1 + ChromaDB ↗
  LangChain, ChromaDB, LLaMA 3.1 8B, Groq API, Python, Streamlit

  Production RAG system — PDF ingestion, vector embedding, semantic retrieval, and multi-turn conversational memory using LLaMA 3.1 8B.

• Medical Summarizer — LLM Document Intelligence ↗
  LLM, NLP, Python, Hugging Face

  LLM-powered system that extracts key information from medical documents and generates structured summaries. Model published on Hugging Face.

• HHS UAC Demand Forecasting Dashboard ↗
  Python, Scikit-learn, Pandas, NumPy, Streamlit Cloud

  End-to-end ML forecasting system on 1,075 days of real government data — 98.2% accuracy on unseen data, live deployed on Streamlit Cloud.

• Dorsal Hand Vein Biometric Authentication ↗
  PyTorch, CNN, Triplet Loss, OpenCV, Python

  Custom CNN with Triplet Loss metric learning for near-infrared biometric verification. Published as peer-reviewed research in IJARCCE (Impact Factor 8.471).

• Facio Music — Real-Time AI Recommendation Agent ↗
  TensorFlow, PyTorch, OpenCV, Python, JavaScript

  Real-time CNN emotion classification (7 classes, 90–95% accuracy) integrated into a live web application for automated music recommendation.

• Symptom Classifier — Clinical NLP System ↗
  Python, Scikit-learn, NLP

  NLP text classification system that maps unstructured patient symptom descriptions to disease/condition categories.

• AI Art Generator — Generative Image Synthesis ↗
  Generative AI, Deep Learning, Python

  Text-to-image generative deep learning model trained from scratch with iterative prompt engineering.

Courses & certifications

• Technology Job Simulation (Coding & Development) · Deloitte / Forage · 2025
• GenAI Powered Data Analytics Job Simulation · Tata / Forage · 2025
• What Is Generative AI? · LinkedIn Learning · 2024
• Machine Learning & AI Foundations: Classification Modeling · LinkedIn Learning · 2024
• Artificial Intelligence Foundations: Machine Learning · LinkedIn Learning · 2024
• Data Analysis with Python and Pandas · LinkedIn Learning · 2024

🗣️ Languages

• English · Fluent
• Hindi · Fluent
• Malayalam · Native