AI/ML Engineer-in-Training with Privacy-First Focus

Aryan Gairola

+91-783******* # ***********@*****.*** ï LinkedIn § GitHub LeetCode

Education

Graphic Era Hill University 2022 – 2026

Bachelor of Technology in Computer Science Dehradun, Uttarakhand Class 12 (CBSE) 2022

Percentage: 91% Dehradun, Uttarakhand

Class 10 (CBSE) 2020

Percentage: 88% Dehradun, Uttarakhand

Technical Skills

Languages: Python, C, C++

Domains: Machine Learning, Deep Learning, NLP, Computer Vision, Generative AI (RAG) Frameworks / Libraries: LangChain, TensorFlow, scikit-learn, Keras, OpenCV, MediaPipe, Flask Tools: Git, GitHub, VS Code, Linux / WSL

Relevant coursework: Data Structures & Algorithms, Competitive Programming, OOP Projects

FinGuard - Privacy-First Fraud RAG 2 Python, CyborgDB, LangChain, Flask

• Built a zero-trust forensic engine using CyborgDB to encrypt vector embeddings and minimize data leakage.

• Engineered an intelligent router to dynamically switch between low-latency chat and deep forensic retrieval.

• Benchmarked the system under high load (concurrent requests), achieving 13 QPS with a 4.2% privacy overhead.

• Developed a real-time dashboard to visualize encryption latency and risk scores for live transaction analysis. Celeb Look-Alike 2 Python, TensorFlow, MediaPipe, Scikit-Learn, Flask

• Fine-tuned a pre-trained ResNet50 architecture on a custom dataset of 1,700+ images.

• Boosted model performance from 45% to 83.4% by unfreezing top layers and applying aggressive Data Augmentation.

• Implemented MediaPipe for robust face alignment and cropping to standardize inputs against pose variations.

• Utilized L2-Normalized Embeddings and Metric Learning principles to enable robust one-shot similarity matching. EviScore 2 Python, Sentence-Transformer, Pydantic, LangChain

• Architected a Safety-First pipeline to extract structured clinical data from multilingual (Hinglish) journals.

• Developed a Semantic N-gram Scorer using Multilingual-BERT to match synonyms and cross-lingual terms.

• Engineered Safety Guardrails via Pydantic and substring checks to eliminate hallucinations and ensure grounding.

• Optimized for restraint using Jaccard Union logic to mathematically penalize redundant or non-grounded output. Achievements

LeetCode: Achieved a peak rating of 1750+ (Top 10% globally). Competitive Programming: Solved 500+ algorithmic problems across various platforms using C++. Leadership: Led a team in a national-level hackathon, coordinating development. Certifications

Google Cloud Computing Foundations 2 NPTEL, IIT Kharagpur Unsupervised Learning, Recommender Systems, Reinforcement Learning 2 Coursera Data Science, Machine Learning, Deep Learning, NLP Bootcamp 2 Udemy

Contact this candidate