Manoj K.

Key Skills

Software

Other

Top Subject Matter

Natural language processing

intended sarcasm detection

Drug discovery

Top Data Types

Text

Image

Medical Dicom

Top Task Types

Fine-tuning

RLHF

Data Collection

Freelancer Overview

Deep-learning Framework to Identify Intended Sarcasm in English (SemEval-2022 Task 6). Brings 6+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Tensorflow and Other. Education includes Master of Technology, Indian Institute of Technology Roorkee (2024) and Master of Science, IISER Bhopal (2024). AI-training focus includes data types such as Computer Code and Programming and labeling workflows including Fine-tuning, RLHF, and Data Collection.

Labeling Experience

Bio Medical Knowledge Base: Leveraging Deep learning Models on One Billion Biomedical Data

OtherData Collection

Consolidated and leveraged one-billion-scale biomedical protein records to construct a large protein-based biomedical knowledge base. Built deep learning models on the aggregated dataset to support tasks such as target identification for disease. The work required organizing heterogeneous inputs from many sources into a unified dataset for downstream AI modeling. • Consolidated data from more than 80 sources into a single protein-based database • Created a database with more than 1 billion records • Built deep learning models for target identification given disease • Designed the biomedical dataset pipeline for large-scale AI training

2023 - 2024

Generative AI and PROTAC concepts to address Sickle Cell Disease

OtherRLHF

Built a proof-of-concept AI framework for PROTAC-related drug discovery for Sickle Cell Disease. Implemented a transformer-based neural network with an embedded reinforcement learning (RL) framework to support model learning and prediction. Trained the model on a large biomedical dataset and also used a gradient-boosted tree model for PROTAC degradation/quality prediction. • Transformer with embedded RL framework (encoder/decoder) • Trained using Tesla K80 GPU on a dataset of over 1 million records • Used LightGBM to predict degradation and assess quality of PROTAC structures • Focused on AI-driven candidate/properties prediction for drug discovery

2023 - 2023

Deep-learning Framework to Identify Intended Sarcasm in English (SemEval-2022 Task 6)

Fine-tuning

Developed and evaluated an intended sarcasm detection deep-learning system for SemEval-2022 Task 6 in English. Performed data augmentation to improve large language model performance and participated in the benchmark evaluation workflow. The project involved training/tuning BERT and RoBERTa-style models and deploying predictions via an API for downstream use. • SemEval-2022 Task 6: Intended Sarcasm Detection In English (SubTask A/B/C) • Used data augmentation techniques to boost accuracy by 5% • Built/deployed a FastAPI-based localhost service for sarcasm percentage prediction • Reported global/subtask ranks from the shared evaluation

2021 - 2022

Education

Work History