Hi, I'm Pothula Akash
LLM Engineer | AI/ML Specialist | Deep Learning Enthusiast
View My WorkAbout Me
I'm an LLM Engineer with a strong foundation in data science and artificial intelligence. I completed my MTech in Data Science from IIT Jodhpur, where I developed a deep passion for working with Large Language Models and advanced AI systems.
My expertise lies in implementing and fine-tuning state-of-the-art language models, building RAG systems, and developing end-to-end machine learning solutions. I enjoy tackling complex problems in natural language processing, computer vision, and deep learning architectures.
With hands-on experience in implementing models from scratch, including GPT-2 and various neural architectures, I'm committed to pushing the boundaries of what's possible with AI while ensuring practical, scalable solutions for real-world applications.
Technical Skills
Large Language Models
LLaMA, Claude, Mistral, GPT Models, Amazon Titan Embed Text V1
Backend & Languages
Python, Express, PyTorch, TensorFlow
Databases
DynamoDB, MySQL, MongoDB
Tools & Cloud
Git, Docker, AWS, CI/CD
Featured Projects
GPT-2 Small (125M Parameters) Implementation
Implemented GPT-2 Small model with 125M parameters from scratch using PyTorch. Validated the implementation by loading pretrained weights from Hugging Face and generating sample predictions. This project demonstrates deep understanding of transformer architectures and attention mechanisms.
AI-Powered Document Question-Answering Chatbot
Built a Retrieval-Augmented Generation (RAG) demo chatbot that can answer questions based on uploaded documents. The system uses advanced embedding models and vector databases to provide accurate, context-aware responses from document content.
NLP Comparative Analysis & Document Classification
Conducted comprehensive analysis comparing NER and POS models from SpaCy, NLTK, and StanfordNER. Implemented document classification using various ML techniques including Naive Bayes, Decision Trees, and Random Forest with TF-IDF vectorization. Achieved detailed performance metrics and confusion matrices for multi-class and binary classification tasks.
Detecting Adverse Drug Events using LSTM
Developed a BiLSTM model for detecting adverse drug events from text data. Implemented custom preprocessing for handling complex entity recognition, used Word2Vec and GloVe embeddings, and achieved 99.2% accuracy with GloVe embeddings. The model effectively identifies drug-adverse effect relationships in clinical text.
Cricket Analysis using Machine Learning
MTP Thesis Project (May 2023 - Nov 2023): Created a unique dataset through web scraping from HowSTAT, performed extensive EDA, and successfully predicted ODI Cricket World Cup final par scores with 10-run accuracy. Implemented multiple ML algorithms achieving 81.2% test accuracy with Random Forest.
Speech Emotion Recognition
Built a speech emotion recognition system using parallel CNN and Transformer encoder architecture. The system recognizes 8 different emotional states from both male and female voices using the RAVDESS dataset. Explored various architectures including Facebook's Hubert model for emotion classification from audio.
Multi-task Classification on CelebA Dataset
Implemented multi-task learning using ResNet18 backbone for classifying 8 facial attributes simultaneously. Used strategic attribute selection for related features, implemented learning rate scheduling, and achieved efficient multi-task performance with shared feature representations.
Teacher-Student Model Implementation
Implemented knowledge distillation using teacher-student architecture with KL divergence loss. Compared performance with and without Exponential Moving Average (EMA) updates. The teacher network uses 12 convolutional layers, demonstrating model compression techniques.
Get In Touch
I'm always interested in discussing new opportunities in AI/ML and LLM engineering.