Abraham
Mathews

AI / ML Researcher

• Designed a multi-agent B2B sales simulation modelling enterprise decision-making units using Perplexity API and LinkedIn data.
• Built a concurrent multi-agent finance Q&A framework using the OpenAI SDK, with agents specialising in stock charting, financial reporting, and economic visualisation.
• Developed SARIMAX forecasting enhanced with LLM-driven sentiment analysis for short-horizon crypto and stock prediction.
• Engineered a PostgreSQL / pgvector RAG system with HNSW indexing for financial news retrieval.

AI / ML Research Intern

• Fine-tuned StyleTTS2 and Coqui VITS models on custom datasets, evaluating naturalness across the resulting systems.
• Built a reproducible voice-cloning pipeline and Gradio interface for non-technical users.

Summer Research Intern

• Designed a fault-tolerant 100 W LLC resonant DC-DC converter for satellite systems, with open-circuit fault detection below 170 µs.

Alien Megastructures in Transit

Developed an end-to-end machine learning pipeline for reconstructing two-dimensional occulting-object shadows from one-dimensional transit light curves. Approximately 200,000 synthetic light curves were generated to train a hybrid 1D-to-2D CNN using TensorFlow and Keras.

The work was presented at the Astronomical Society of India meeting at IISc Bengaluru, and at the Interstellar Frontiers conference in Perth.

Thrust Vector Control

Prototyped a servo-driven gimbal thrust-vector-control system for a solid-propellant model rocket using an Arduino Nano and MPU6050 flight controller. The system was validated through static hold-down tests and MATLAB / Simulink 3-DOF modelling.

Fault-Tolerant LLC Converters

Investigated open-circuit switch fault tolerance in half-bridge LLC resonant converters intended for satellite applications.