Srijan S.

Key Skills

Software

Other

Top Subject Matter

Privacy-preserving multimedia security (Secret Image Sharing)

Lab report evaluation and standardized assessment (data quality and correctness)

Top Data Types

Text

Image

Top Task Types

Fine-tuning

Freelancer Overview

Backed by a strong academic foundation in advanced computing, my experience in AI training data and data labeling centers on preparing high-quality, secure datasets for complex machine learning models. Through specialized coursework in Topics in Deep Learning and Information Theory, Inference, and Learning Algorithms, I have developed a deep understanding of data preprocessing, feature extraction, and the critical role that precise annotations play in model convergence and accuracy. I am highly skilled at data curation, quality assurance, and handling complex labeling schemas for neural networks, ensuring that training pipelines are fed with clean, mathematically sound ground-truth data. What sets me apart is my ability to scale and secure data operations. Leveraging my background in Distributed and Parallel Computing, I understand how to manage large-scale data pipelines efficiently, while my training in Software and Cyber Security ensures that I approach data handling with strict adherence to privacy, data integrity, and secure storage protocols. This unique combination of deep learning theory, distributed systems engineering, and cybersecurity allows me to not only label data accurately but also optimize the underlying pipeline for enterprise-grade AI training

Labeling Experience

Context-Aware PII and Entity Annotation for LLM Privacy Gateways

TextEntity (NER) Classification

Engineered and curated a high-quality, fine-grained NER dataset designed to train token-classification models (such as GLiNER and Qwen architectures) for zero-trust AI prompt sanitization. The core objective was the precise identification, annotation, and redaction of complex Named Entities and Personally Identifiable Information (PII) embedded within diverse, unstructured user prompts. Key Responsibilities & Methodology: Schema Design & Annotation: Developed and executed a rigorous labeling schema spanning standard entities (Names, Locations, Organizations) and complex technical entities (API keys, cryptographic hashes, IP addresses, proprietary source code snippets, and financial metadata). Context-Aware Disambiguation: Resolved edge cases where overlapping entities existed, ensuring the model could differentiate between benign technical prose and sensitive data leak risks based on surrounding context. Quality Assurance & Benchmarking: Conducted iterative data auditing and conflict resolution to maintain a high Inter-Annotator Agreement (IAA) score. Used the labeled dataset to benchmark downstream token-classification models, ensuring robust generalization and minimal false-positive rates in automated data redaction.

2026 - 2026

Education

Work History