Skills

• Tools & Technologies: Python, SQL & Cloud (AWS)
• Libraries & Frameworks: PyTorch, Scikit‑learn, Pandas, Numpy, XG‑Boost, LightGBM, NLTK, SpaCy, Gradio, Spark, Hadoop, Transformers
• Algorithms & Concepts: Regression (Linear/Logistic), Decision Trees, Random Forest, Gradient Boosting Machine, XGBoost, Clustering, PCA, Neural Networks, Deep Learning, Large Language Models, Prompt Engineering, Retrieval Augmented Generation
• Others: LaTeX, MS Office (Word, Excel, PowerPoint), Confluence, Git, Jira, Kanban, Mural
• Expertise: Natural Language Processing / Understanding / Generation
• Domains: FinTech (Financial Services + Technology), Consumer Internet based Products & Customer Analytics

Selected Projects

Voice of Customers

Experimented with Large Language Models (T5, Falcon, MPT, Open‑Laama). Created Bi‑LSTM, T5, LED & BART based models for summarizing & extracting multiple themes from call transcripts. Used Agglomerative Hierarchical Clustering recursively to group similar themes. Designed the annotation job using Appen and administered a team of 98 annotators.
Applications:
i) Automated short note generation for calls,
ii) Comprehending reasons behind high call volumes,
iii) Featurization of textual interaction data.
Impact:
i) Issues related to login were identified & resolved leading to 10% increase in customer satisfaction score
ii) Workflow for different processes were updated leading to increase in Net Promoter Score and Customer Ease Score by 18 points & 27% respectively
iii) Competent solution to address the student debt crisis was created which lead to 60% increase in enrollment, 59K yearly payments & $200M+ being paid to the participants.

Relevant Publications:

• “AUTOMATED ANALYSIS OF CUSTOMER INTERACTION TEXT TO GENERATE CUSTOMER INTENT INFORMATION AND HIERARCHY OF CUSTOMER ISSUES”, US Patent publication No. US20230112369A1

Retrieval Augmented Generation (RAG) on Call Transcripts

Experimented with Retrieval Augmented Generation (RAG) to answer business queries from call transcripts. Worked on various chunking and denoising strategies. Explored several frameworks like LangChain, LlamaIndex, FAISS, Chroma DB, etc.

Inclusive Investing

Making the investment process more inclusive so that even the economically lower strata of the society can avail financial services.

Topics:

• Improving readability of financial texts
• Improving reach & engagment of financial social media posts

Relevant Publications

• “FinRAD: Financial Readability Assessment Dataset - 13,000+ Definitions of Financial Terms for Measuring Readability” in FNP@LREC-2022
• “Generator-Guided Crowd Reaction Assessment” in TheWebConf (WWW) 2024

Improved Investing

Improving the journey of investments

Topics:

• Extracting hypernyms of financial terms
• Extracting relationship between financial entities

Relevant Publications:

• “Term Expansion and FinBERT fine-tuning for Hypernym and Synonym Ranking of Financial Terms” in FinNLP@IJCAI 2021
• “Learning to Rank Hypernyms of Financial Terms using Semantic Textual Similarity” in SN Computer Science (Springer) 2023
• “The Mask One At a Time Framework for Detecting the Relationship between Financial Entities” in FIRE 2023

Impactful (Green) Investing

Considering environmental aspects while investing

Topics:

• Classifying a financail text as Sustainable or Unsustainable
• Detecting Environmental, Social and Governance (ESG) Issues from financial texts
• Identifying ESG impact type
• Identifying ESG impact duration

Relevant Publications:

• “Ranking Environment, Social And Governance Related Concepts And Assessing Sustainability Aspects Of Financial Texts” in IJCAI-ECAI 2022
• “A low resource framework for Multi-lingual ESG Impact Type Identification” in FinNLP@IJCNLP-AACL 2023

Informed Investing

Keeping the investors informed and helping them to make data driven decisions

Topics:

• Detecting exaggerated and in-claim numerals from Financial Texts
• Evaluating the Rationals of Amateur Investors
• Evaluating the effect of Social Media Posts by Executives on Stock Prices
• Fine-grained Argument Understanding in Financial Texts

Relevant Publications:

• “LIPI at the NTCIR-16 FinNum-3 Task: Ensembling transformer based models to detect in-claim numerals in Financial Conversation” in NTCIR-16 2022
• “Evaluating Impact of Social Media Posts by Executives on Stock Prices” in FIRE 2022
• “LIPI at the FinNLP-2022 ERAI Task: Ensembling Sentence Transformers for Assessing Maximum Possible Profit and Loss from Online Financial Posts” in FinNLP@EMNLP 2022
• “LIPI at the NTCIR-17 FinArg-1 Task: Using Pre-trained Language Models for Comprehending Financial Arguments” in NTCIR-17 2023

Indic Investing

Helping Indians to manage their wealth

Relevant Publications:

• “Financial Argument Analysis in Bengali” in FIRE 2023
• “IndicFinNLP: Financial Natural Language Processing for Indian Languages” in LREC-COLING 2024

FinNLP tools

Open sourcing tools for analysing financial texts

Relevant Publications:

• “FinRead: A Transfer Learning Based Tool to Assess Readability of Definitions of Financial Terms” in ICON-2021
• “Fincat: Financial numeral claim analysis tool” in FinWeb@WWW 2022
• “Fincat-2: An enhanced Financial Numeral Claim Analysis Tool” in Software Impacts (Elsevier) 2022
• “FLUEnT: Financial Language Understandability Enhancement Toolkit” in CODS-COMAD 2023