Shivam J.

Key Skills

Software

Toloka

OneForma

Mindrift

Micro1

Mercor

Top Subject Matter

Named Entity Recognition (NER) and multi-modal classification

ASR post-processing and NLP candidate ranking

Legal Services & Contract Review

Top Data Types

Text

Document

Image

Top Task Types

Entity (NER) Classification

Text Generation

Bounding Box

Classification

RLHF

Cuboid

Freelancer Overview

Multi-Modal Natural Language Processing Engine (NLP) — Spring 2023. Brings 7+ years of professional experience across legal operations, contract review, compliance, and structured analysis. Core strengths include PyTorch, SciKit Learn, and pandas. Education includes Bachelor and Master of Technology, Indian Institute of Technology Delhi (2020) and High School Diploma, Amity International School (2020). AI-training focus includes data types such as Text, Computer Code, and Programming and labeling workflows including Entity (NER) Classification and Text Generation.

Labeling Experience

AI-Powered Programmable Communications Gateway — Spring 2025

Text Generation

Architected an ASR post-processing engine to dynamically parse, correct, and rewrite noisy spoken transcripts into structured text formats such as emails, currency notation, and numbers. Implemented an optimized multi-candidate scoring pipeline using an INT8 quantized DistilBERT Masked-LM executed in ONNX Runtime to reduce tail latency. Engineered batched pseudo-likelihood ranking to score candidate token sequences efficiently and bypass CPU bottlenecks. • Built transcript parsing and rewrite rules for structured outputs • Ran vectorized multi-candidate scoring with ONNX Runtime and DistilBERT • Implemented batched pseudo-likelihood token sequence ranking • Deployed regex-based threshold short-circuiting for high-confidence matches

2025 - 2025

Multi-Modal Natural Language Processing Engine (NLP) — Spring 2023

TextEntity (NER) Classification

Built sequential network evaluation modules to track precision and recall performance drops across RNN, LSTM, and Bi-LSTM architectures for targeted Named Entity Recognition (NER). Engineered deep multi-modal classification modules using multiple transformer-based baselines to analyze complex, unstructured graphic media distributions. Focused on model evaluation and performance monitoring to support supervised NLP training workflows. • Evaluated NER model variants across architectures • Measured precision/recall changes to diagnose model regressions • Trained/validated multi-modal classification ensembles for unstructured media • Used performance tracking outputs to guide iteration and tuning

2023 - 2023

Education

Work History