I Engineer Intelligence into Digital Ecosystems.

Kerala, India

I'm an AI/ML Engineer specializing in deep learning, computer vision, and scalable data pipelines. I bridge the gap between complex mathematical models and real-world execution.

Profile picture

Professional Journey

2025 - 2026

AI/ML Intern

International Centre for Free and Open Source Software (ICFOSS)

Developed DhritiOCR, a comprehensive text extraction Web UI and ML pipeline. Achieved state-of-the-art performance for an 80M-parameter class by fine-tuning a custom recognition model for Malayalam (Indic language) that overcomes the inherent structural challenges of low-resource scripts.

2025 - 2026
2025

Machine Learning Intern

International Centre for Free and Open Source Software (ICFOSS)

Built a custom ML audio pipeline (ADAPT) for automated speaker diarization, splitting, and transcription. Architected core ML modules for advanced grapheme-to-phoneme conversion using custom Malayalam phonetic analyzers.

2025
2024 - 2025

Open IoT Student Ambassador

International Centre for Free and Open Source Software (ICFOSS)

Maintained and troubleshot deployed IoT sensor networks with a focus on LoRaWAN weather stations. Built a practical foundation in physical computing, low-level hardware diagnostics and edge device management.

Tech Stack

The core frameworks, libraries, and tools I leverage to build scalable machine learning pipelines and secure architectures.

Python
PyTorch
CUDA
ROCm
ONNX
vLLM
llama.cpp
OpenCV
PaddlePaddle
Transformers
Docker
FastAPI

Featured Projects

A selection of my most impactful machine learning architectures and data-driven solutions.

View All Repositories
Screenshot of DhritiOCR: Document level OCR System for Malayalam
Text Extraction & Recognition

DhritiOCR: Document level OCR System for Malayalam

Developed DhritiOCR, a comprehensive text extraction Web UI and ML pipeline. Achieved state-of-the-art performance for an 80M-parameter class by fine-tuning a custom recognition model for Malayalam (Indic language) that overcomes the inherent structural challenges of low-resource scripts.

PaddlePaddle
CUDA
ONNX
OpenCV
FastAPI
Learn More
Screenshot of ADAPT: Audio Data Annotation and Preprocessing Tool
Audio Processing & ML Pipelines

ADAPT: Audio Data Annotation and Preprocessing Tool

A high-performance CLI dataset generation pipeline for TTS and ASR. Ingests raw audio or YouTube streams and outputs clean, diarized, transcribed data, leveraging CUDA and ROCm for maximum hardware acceleration.

Python
PyTorch
Transformers
CUDA
ROCm
Learn More
Screenshot of Clara : Cybersecurity based LLM for Anomaly and Risk Assessor
Cybersecurity & LLMs

Clara : Cybersecurity based LLM for Anomaly and Risk Assessor

An intelligent security partner designed to accelerate vulnerability detection through advanced code comprehension. Clara is an LLM post-trained on a refined PrimeVul dataset, optimized to identify complex security weaknesses that traditional static analysis tools often overlook, ensuring a robust and secure development lifecycle.

Python
PyTorch
Transformers
llama.cpp
vLLM
Learn More

Research Studies

Applied ML Research

DhritiOCR: Lightweight OCR Pipeline for Indic Languages

A custom document level OCR system utilizing the PaddlePaddle framework. The pipeline features models fine-tuned to achieve SOTA text extraction performance for Malayalam-English scripts alongside native structural preservation for wired tables.

Coming Soon
Academic Research

Module-FHE: Vectors are All You Need

An investigation into the algebraic structure of Module-LWE as a foundational framework for constructing practical Fully Homomorphic Encryption (FHE) schemes. The inherent vector-like structure of modules over polynomial rings provides a natural and powerful paradigm for designing highly parallel and efficient homomorphic operations.

Coming Soon