ADITYA DUBEY

LLM-Based Macroeconomic Policy Agents via Reinforcement Learning with DSGE Simulators: Researching the use of LLMs combined with RL and DSGE simulators to develop macroeconomic policy agents capable of optimizing interventions based on simulated economic outcomes. Addressed gaps in static economic models by designing a framework for reward-driven, real-time policy feedback and exploring applications in forecasting, scenario analysis, and academic theory testing.

• Utilized web parsing, heuristic approaches, NLP techniques, and LLM-based classification to develop data labeling workflows with integrated confidence scoring.
• Applied fine-tuning of Sentence Transformers through knowledge distillation to improve multilingual sentence embeddings, leveraging research-backed methods for cross-lingual alignment.
• Explored the use of Siamese Neural Networks for short-text similarity and one-shot learning, particularly for tasks involving minimal-context labels, informed by domain literature.
• Conducted semantic similarity evaluations using fine-tuned models to support label alignment and categorization across multilingual datasets.
• Working on developing Queries for context building for a Retrieval Augmented Generation(RAG) model, using Natural Language Processing(NLP).

• A SOP(Study Project) to maintain the backend development of the Project Backend-Enabled Electronic Media. SOP under Dr. Vinay Chamola.

Developing a context identification module for Retrieval-Augmented Generation (RAG) applications specifically trained for the Energy domain using Entity Recognition (NER) and NLP techniques to enhance RAG model performance.

• Worked in Borrow App Development: Designed and implemented the authentication and authorization system, managed database setup and AWS hosting.
• Part of the Backend Vertical Team, working on numerous projects, recruitments etc.