Hi, I'm Akhil Theerthala

Exploring the Gap between Human Intent & AI.

"We shape our tools and thereafter our tools shape us."

But this cycle breaks if our tools don't understand us. I explore the gap between Human Intent & AI, focusing on models that adapt to how people actually think and work, rather than forcing us to adapt to the machine.

2.7 Yrs. Work Experience
6.7k+ HF Downloads
30k+ Community Model Downloads
Akhil Theerthala

About Me

I am a Senior Data Scientist with over 2 years of FinTech experience, specializing in data-centric methods for resource-efficient LLMs. My work focuses on developing generalized financial reasoning models and validating their performance through rigorous ablation studies.

I have a proven track record of community impact through open-source personal finance models, with community-adapted versions attracting over 31k+ downloads. In my professional roles, I've achieved significant business results, including a 97.5% reduction in document processing time and a 27.6% improvement in table detection performance.

Work Experience

Mar 2025 - Present

Senior Data Scientist

Perfios Software Solutions

  • Benchmarked Vision Language Models for table structure recognition, yielding a 0.85 TEDS score on bank statements.
  • Engineered a document legibility framework that reduced analysis time for documents with unreadable text by 20%.
  • Designed and prototyped multiple agentic workflows for underwriting, establishing a proof-of-concept for new products.
Jun 2023 - Mar 2025

Data Scientist

Perfios Software Solutions

  • Reduced document processing time by 97.5% (from 8s to 200ms) by implementing and optimizing a multimodal classifier.
  • Boosted table detection performance by 27.6% by engineering a new module and evaluating multiple model architectures.
Aug 2022-Apr 2023

UG Student Researcher

VGSOM, IIT Kharagpur

Investigating the impact of social media posts on crowdfunding projects, under the guidance of Dr. Swagato Chatterjee.

  • Identifying the characteristics that inspire social media users to participate in crowdfunding campaigns
  • Identifying the relationship between social media engagement and the amount of funds raised for the initiative
  • Identifying the connection between the funds raised and the uncovered motivating factors
May 2022 - Jul 2022

Data Science Intern

Perfios Software Solutions

During my three-month internship, I worked on projects such as evaluating business intelligence tools and recommending banking products to current account holders.

Skills

ML & Deep Learning

PyTorch HuggingFace (Transformers, TRL) TensorFlow Scikit-learn NLTK OpenCV

Data Science

Pandas NumPy Matplotlib Seaborn Plotly Scrapy

Tools

Docker Git Tensorboard Selenium

Featured Projects

Kuvera: Personal Finance LLM

Engineered a cost-effective pipeline to generate a 20k-sample Personal Finance Chain-of-Thought (CoT) dataset. Fine-tuned 8B/14B parameter LLMs for efficient single-GPU deployment.

1.4k+ Dataset Downloads 500+ Model Downloads
View on HuggingFace

Crowdfunding Engagement Drivers

Engineered an NLP pipeline to extract features from Facebook posts. Achieved a 0.82 F1 score with an ensemble model.

NLP Ensemble Learning
View Code

comic-genesis

Comic book creator using Gemini Nano.

GenAI Gemini
View Code

Achievements & Education

Achievements

  • 1st Place in the Reasoning Dataset Creation Competition (HuggingFace, Together.ai, Bespoke Labs).
  • Finalist in the Perfios IdeaFest for pitching an AI-powered healthcare product.

Education

Indian Institute of Technology, Kharagpur

B. Tech, Aerospace Engineering (2019 – 2023)

Coursework: Graphical & Generative Modelling, Dependable AI-ML, DL Foundations.

Publications

A Data-Centric Framework for Training Behaviour-Aware Personal Finance Language Models

Akhil Theerthala (2025) • FinNLP Workshop at EMNLP

Latest Articles

Creating a Reasoning Dataset with No Budget

Apr 7

Read on Medium

From Training Language Models to Training DeepSeek-R1

Feb 17

Read on Medium

7 Practical PyTorch Tips

Feb 10

Read on Medium