Overview

Position: AI Research Intern
Company: MyFi by TIFIN
Duration: August 2024 - December 2024
Location: Remote

Following my Berlin internship, I joined MyFi by TIFIN as an AI Research Intern, where I spearheaded the multilingual expansion of their financial advisory mobile application. This project was particularly challenging as it required bridging cutting-edge NLP research with practical production constraints in the FinTech domain.

Key Projects

Multilingual LLM Evaluation

  • Conducted comprehensive comparative analyses of state-of-the-art multilingual large language models
  • Evaluated models including Sarvam-2b, JAIS, and fine-tuned Llama variants specifically for Hindi and Indic languages
  • Systematically benchmarked different approaches—from training from scratch to fine-tuning existing models
  • Determined optimal strategy for MyFi’s unique requirements balancing performance, latency, and linguistic accuracy

Hinglish Challenge Solution

  • Solved the Hinglish challenge—enabling natural handling of colloquial code-mixing of Hindi and English prevalent in Indian conversations
  • Developed sophisticated language detection capabilities for seamless code-switching
  • Implemented careful prompt engineering to maintain professional tone necessary for financial advice while supporting informal conversational patterns
  • Bridged the gap between casual user communication and professional financial advisory standards

Financial Information Parser Development

  • Developed sophisticated parser system capable of extracting structured financial information from natural language queries
  • Created synthetic datasets for training specialized financial parsers
  • Fine-tuned multiple Qwen2.5 models for different financial information extraction tasks
  • Published several specialized parsers on Hugging Face, making them available to the broader community
  • Designed effective training data for an evolving financial ontology with research creativity and engineering rigor

Technical Achievements

Model Development

  • Research and Implementation: Balanced research innovation with production deployment requirements
  • Multilingual Expertise: Deep understanding of Indic language processing challenges and solutions
  • Fine-tuning Optimization: Optimized models for production deployment while maintaining accuracy

Production Impact

  • Parser Integration: Developed parsers became foundational components of MyFi’s query understanding infrastructure
  • Immediate Practical Impact: Delivered research outcomes with direct application to production systems
  • Community Contribution: Open-sourced specialized models for broader financial NLP community

Skills Developed

  • Multilingual NLP: Advanced techniques for handling code-mixing and language detection
  • Financial Domain Knowledge: Understanding of financial advisory requirements and user interaction patterns
  • Research to Production: Translating research innovations into deployable production systems
  • Synthetic Data Generation: Creating effective training datasets for specialized domains

Impact

This internship demonstrated my ability to tackle complex, domain-specific NLP challenges while maintaining focus on practical deployment constraints. The work required both deep technical expertise and understanding of cultural and linguistic nuances in financial communication.

The experience reinforced my passion for applied AI research and prepared me for the transition to full-time engineering roles where research insights must be translated into robust, scalable systems.