Shaping Intelligence: The ACM Gen AI Team

Introducing the newest addition to ACM BPHC: The Gen AI Team, our hub for applied artificial intelligence. We explore the foundations of AI by building models from the ground up to solve complex, real-world challenges. We don’t just use AI; we engineer it. Our focus is on engineering intelligent systems that can see, understand, translate, and create.

What We Build

Our projects tackle diverse problems in Natural Language Processing (NLP) and Computer Vision (CV). We emphasize a deep understanding of the fundamentals by building, training, and evaluating our own neural network architectures.

1. Custom Sequence-to-Sequence Translator

This project demystifies machine translation by building a complete language translator from scratch, without relying on pre-trained models.

A custom LSTM Encoder-Decoder architecture to understand the flow of information.
Handles different source and target language vocabularies.
Implements teacher forcing for more stable and efficient training.
Core Concepts: Sequence-to-Sequence (Seq2Seq), LSTMs, Encoder-Decoder Frameworks, NLP.

2. Deep Learning for Structural Defect Detection

An automated system that uses computer vision to identify cracks and defects in materials, making infrastructure inspection faster and more reliable.

A custom-built Convolutional Neural Network (CNN) for high-accuracy image classification.
Uses data augmentation techniques to create a robust and generalized model.
Classifies images to pinpoint potential structural integrity issues.
Core Concepts: CNNs, Computer Vision, Image Classification, Supervised Learning.

3. AI-Powered Image Enhancer Suite

A versatile toolkit of specialized deep learning models that intelligently improve and artistically transform images.

Features five distinct models: B&W image colorizer, denoiser, resolution upscaler, brightness/contrast corrector, and neural style transfer.
A smart meta-model analyzes the input image to recommend the best enhancement tool.
Each model is trained from scratch on a specialized dataset.
Core Concepts: Autoencoders, U-Nets, Super-Resolution, Style Transfer.

4. Autonomous Driving Perception System

A multi-model system that serves as the "eyes" for an autonomous vehicle, analyzing video streams to identify and report road hazards in real-time.

Five parallel CNN models for detecting pedestrians, vehicles, traffic signs, and road conditions.
Analyzes video frame-by-frame and determines the spatial location of obstructions (left, right, straight).
An LLM translates the model's findings into clear, human-readable warnings.
Core Concepts: Object Detection, Multi-Modal AI, Real-Time Inference, LLM Integration.

How We Build

We use foundational frameworks and libraries to gain a deep understanding of AI model development from first principles.

Frameworks: PyTorch, TensorFlow
Core Architectures: Foundational Neural Networks, Sequential Models, Advanced Computer Vision Architectures
Domains: Natural Language Processing (NLP), Computer Vision (CV), Agentic AI, Retrieval Augmented Generation (RAG), Supervised & Unsupervised Learning
Libraries: LangChain, NumPy, Pandas, Scikit-learn, OpenCV

Why Join the Gen AI Team?

Joining the ACM Gen AI Team means building practical skills in a rapidly advancing field. It's an opportunity to move past theory and apply AI concepts to tangible projects. You will:

Build AI from the ground up, going beyond APIs to truly understand how models work.
Gain hands-on experience in a high-growth area of technology.
Solve diverse, impactful problems across NLP, computer vision, and autonomous systems.
Collaborate with a driven team of thinkers and builders who share your passion.
Develop a portfolio that showcases deep, foundational AI expertise.

If you are motivated to build powerful, intelligent systems and want to develop a strong technical foundation in AI, this is the team for you.