GAURAV BHARDWAJ

AI Trainer

Jamshedpur, IN.

About

Developed detailed annotations for over 500 data points used in training AI models, identifying key areas for optimization that reduced output inaccuracies by 30%. Streamlined the AI-generated content review process through structured feedback loops, improving response clarity and consistency across automated interactions. Collaborated with cross-functional teams to ensure model outcomes aligned with business requirements, contributing actionable data insights for the business team.

Work

Outlier

|

AI Trainer

Summary

Developed detailed annotations for over 500 data points used in training AI models, identifying key areas for optimization that reduced output inaccuracies by 30%. Streamlined the AI-generated content review process through structured feedback loops, improving response clarity and consistency across automated interactions. Collaborated with cross-functional teams to ensure model outcomes aligned with business requirements, contributing actionable data insights for the business team.

Highlights

Developed detailed annotations for over 500 data points used in training AI models, identifying key areas for optimization that reduced output inaccuracies by 30%.

Streamlined the AI-generated content review process through structured feedback loops, improving response clarity and consistency across automated interactions.

Collaborated with cross-functional teams to ensure model outcomes aligned with business requirements, contributing actionable data insights for the business team.

Education

INDIAN INSTITUTE OF TECHNOLOGY, DHANBAD

Bachelor of Technology (B. Tech) Electrical Engineering

RAJENDRA VIDYALAYA, JAMSHEDPUR

→
Class-12 ICSE Grade: 93.75 %

RAJENDRA VIDYALAYA, JAMSHEDPUR

→
Class-10 ICSE Grade: 95.8 %

Awards

Ranked 14th out of 5000+ participants data science competition by IITG

achieving significant performance improvements in an Faster-RCNN, trained on 1,800+ HD images to detect automobile injuries (68%). Project Link

Analyzed 10K data samples adressing loan defaults in quant research program at JPMorgan Chase

crafted effective strategies that directly improved risk prediction accuracy by enhancing existing models.

Global rank 142 in CodeChef Starters 66.

Global rank 1740 in Google Kickstart Round H 2022.

All India rank 5963 in JEE Advanced 2021.

Skills

Languages

C, C++, Python, Swift, SQL, Java(basics).

Technologies/Frameworks/Libraries

GCP, AWS, NLP, Computer Vision, Sagemaker, vector.ai, Pytorch, LangChain, Hugging Face, Matlab, Postgres, Tableau, streamlit, MongoDB, GIT.

Others

Time Series Analysis, OOPS, GPT, RAG, Terraform, Prompt Engineering, bash scripting.

Projects

Sentiment Analyser

Summary

Developed and implemented a robust sentiment analysis model using BERT and NLTK, processing data from over 60,000 tweets to provide actionable insights that influenced trading strategies for 25 high-profile stocks. Integrated advanced data-driven sentiment signals into portfolio risk analytics, enhancing the predictive accuracy and enabling more informed investment decisions across various asset classes.

STEM Solver

Summary

Developed OpenELM-3B model for STEM/math tasks, achieving 2.36% higher accuracy than OLMo with 2× fewer pre-training tokens. Strong on ARC-c(42.24), MMLU(26.76), Hellaswag (73.28) benchmarks. Enhanced base model performance through targeted fine-tuning using LoRA, synthetic data and PRMs with Metal optimization and quantization, deploying final model via FastAPI for inference.

Auto Insurance Claim Resolution

Summary

Implemented a multi-agent workflow to review insurance claims by analysing policy documents, damaged photos, and repair estimates, cutting manual review time by 40%. Trained the system using 500+ past claims data to improve accuracy. Integrated it with existing claims databases, allowing it to process 90% of routine cases without human input.