Ayush Tanwar profile photo 1 on Tal

Ayush Tanwar

ML Engineer | GenAI, LLM Systems, RAG & LLMOps | Fraud ML

Gurugram

4+ years of work experience

American Express
May 20251 yr 4 mosPresent

Gurugram · Hybrid

- Built a transcript-driven fraud risk modeling system to capture behavioral and linguistic signals not present in transaction data. - Engineered large-scale NLP features and trained an XGBoost model, achieving ~25% incremental fraud coverage. - Trained and evaluated a FinBERT classifier to identify fraud-related semantic and intent signals in call transcripts. - Developing a GenAI based reasoning and scoring pipeline, grounded in retrieved domain knowledge from curated reference documents and validated through structured ofline benchmarking and LLM-based evaluation for phone-based identity takeover.
American Express
Sep 20231 yr 9 mosMay 2025

Gurugram · Hybrid

-Twin Tower Transformer to improve Fraud Coverage on underperforming segments:Transformer Model to improve Fraud coverage on underperforming segments leveraging Specific Attention. Ideated & Evaluated a Twin-Tower Transformer architecture to capture temporal and intra-transactional patterns.Improved the Recall Score at riskiest 20bps by 1.2% with an expected annual benefit of $6MM.Fraud Indicator -Meta Learner: MAML-Optimized RNN for Regional Sequence Adaptation: - Developed a MAML-driven RNN to learn regional and regime-shifting sequence patterns, boosting model coverage by 10%. - Built a meta-learning pipeline with fast adaptation loops enabling reliable performance on new regional/time-variant tasks. - Delivered a PyTorch-based system with stable gradient updates and fast-weight adaptation for improved temporal generalization.
Spry Movement App
Jun 20223 mosAug 2022

- Worked on various areas of development focusing on GPT3 based NLP models and fine-tuning large-scale language models. - Successfully developed a web app that facilitates Automated AI content writing for Blogs and clinic descriptions. - The web app is consistently being used for blog writing and Search Engine Optimization
Lam Research
May 20223 mosJul 2022

Bengaluru, Karnataka, India

- Developed an analytical model to calculate the stresses developed in deposited layers on semiconductor wafers. - Improved the Stoney's equation by performing simulations for various parameters in ABAQUS and achieved an accuracy of 90%. - Automatized the whole simulation process using Python and created a web app Stress Calculator to reduce the modeling time
Institute of Actuaries of India
Dec 20219 mosAug 2022

Qualifed ACET and became a student member of the Institute of Actuaries of India (IAI, Ministry of Finance)
DRDO, Ministry of Defence, Govt. of India
May 20214 mosAug 2021

Roorkee, India

- Worked on a project of national importance with the Department of Science and Technology (DST), INDO-SRILANKAN Research Project - Developed various origami-based toroidal patterns and performed the deployment simulations using ABAQUS-CAE - Conducted detailed stress-strain studies using ABAQUS and MATLAB to find the optimal parameters for the satellite deployment
GitHub
@Flakes342

Ayush’s AI Stack

antigravity
antigravity
claudecode
claudecode

Education

Indian Institute of Technology, Delhi

Bachelor of Technology - BTech, Mechanical Engineering

Indian Institute of Technology, Delhi

20194 yrs2023

Heera Lal Public School

20172 yrs2019

Join Ayush and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Ayush Tanwar on Tal

Headline

ML Engineer | GenAI, LLM Systems, RAG & LLMOps | Fraud ML

About

Machine learning engineer working across applied machine learning, data science, and AI research. At American Express, I develop fraud and risk models using large-scale real-world data, with a focus on understanding model behavior, limitations, and tradeoffs rather than emphasizing infrastructure alone. My interests center on robustness, interpretability, error analysis, and representation learning, as well as emerging work in LLMs, retrieval-augmented methods, and multi-agent systems. I aim to bridge practical impact with research-driven rigor.

Location

Gurugram

AI stack

antigravity

claudecode

Work experience

Senior Data Scientist - Gen AI & Fraud Analytics at American Express, May 2025 - Present · 1 yr 4 mos, Gurugram · Hybrid, - Built a transcript-driven fraud risk modeling system to capture behavioral and linguistic signals not present in transaction data. - Engineered large-scale NLP features and trained an XGBoost model, achieving ~25% incremental fraud coverage. - Trained and evaluated a FinBERT classifier to identify fraud-related semantic and intent signals in call transcripts. - Developing a GenAI based reasoning and scoring pipeline, grounded in retrieved domain knowledge from curated reference documents and validated through structured ofline benchmarking and LLM-based evaluation for phone-based identity takeover.

Data Scientist - Fraud Analytics at American Express, Sep 2023 - May 2025 · 1 yr 9 mos, Gurugram · Hybrid, -Twin Tower Transformer to improve Fraud Coverage on underperforming segments:Transformer Model to improve Fraud coverage on underperforming segments leveraging Specific Attention. Ideated & Evaluated a Twin-Tower Transformer architecture to capture temporal and intra-transactional patterns.Improved the Recall Score at riskiest 20bps by 1.2% with an expected annual benefit of $6MM.Fraud Indicator -Meta Learner: MAML-Optimized RNN for Regional Sequence Adaptation: - Developed a MAML-driven RNN to learn regional and regime-shifting sequence patterns, boosting model coverage by 10%. - Built a meta-learning pipeline with fast adaptation loops enabling reliable performance on new regional/time-variant tasks. - Delivered a PyTorch-based system with stable gradient updates and fast-weight adaptation for improved temporal generalization.

Data Science Intern at Spry Movement App, Jun 2022 - Aug 2022 · 3 mos, - Worked on various areas of development focusing on GPT3 based NLP models and fine-tuning large-scale language models. - Successfully developed a web app that facilitates Automated AI content writing for Blogs and clinic descriptions. - The web app is consistently being used for blog writing and Search Engine Optimization

Product Development Intern at Lam Research, May 2022 - Jul 2022 · 3 mos, Bengaluru, Karnataka, India, - Developed an analytical model to calculate the stresses developed in deposited layers on semiconductor wafers. - Improved the Stoney's equation by performing simulations for various parameters in ABAQUS and achieved an accuracy of 90%. - Automatized the whole simulation process using Python and created a web app Stress Calculator to reduce the modeling time

Student Member at Institute of Actuaries of India, Dec 2021 - Aug 2022 · 9 mos, Qualifed ACET and became a student member of the Institute of Actuaries of India (IAI, Ministry of Finance)

CAE/FEA Analyst Intern at DRDO, Ministry of Defence, Govt. of India, May 2021 - Aug 2021 · 4 mos, Roorkee, India, - Worked on a project of national importance with the Department of Science and Technology (DST), INDO-SRILANKAN Research Project - Developed various origami-based toroidal patterns and performed the deployment simulations using ABAQUS-CAE - Conducted detailed stress-strain studies using ABAQUS and MATLAB to find the optimal parameters for the satellite deployment

Education

Bachelor of Technology - BTech, Mechanical Engineering, Indian Institute of Technology, Delhi

Heera Lal Public School

Public links

github: https://github.com/Flakes342