Akshat Tripathi
AI & ML engineer building production-ready impact models
AI Invoice Details Extractor
An AI-powered invoice processing system designed to automatically extract structured information from invoices, reducing the need for manual data entry and making document processing faster and more scalable.
What I Built
I developed an end-to-end AI invoice information extraction pipeline that processes invoice documents and identifies important fields such as:
Invoice number
Invoice date
Vendor/supplier details
Customer details
Billing and shipping information
Tax information
Subtotal and total amount
Line-item details
Product/service descriptions
The project uses deep learning and document understanding techniques to understand both the textual content and spatial layout of invoice documents.
Dataset & Preprocessing
I collected invoice/document data from Hugging Face and built a preprocessing pipeline to prepare the documents for model training.
The preprocessing workflow included:
Dataset cleaning and filtering
Document/image processing
Text extraction and normalization
Annotation preparation
Layout information processing
Tokenization
Training/validation dataset preparation
Special attention was given to preserving the relationship between text and its position within the document, which is important for understanding complex invoice layouts.
Model Development
For the extraction model, I worked with LayoutLMv3 using PyTorch, leveraging its ability to combine:
Text + Visual Information + Document Layout
This allows the model to understand not only what information appears on an invoice, but also where it appears and how different elements are related.
End-to-End Pipeline
Invoice → Preprocessing → Layout/Text Understanding → AI Model → Field Extraction → Structured Data
The extracted information can then be converted into structured formats such as JSON or tabular data, making it suitable for integration into accounting, ERP, finance, or document-processing workflows.
Key Technologies
Python • PyTorch • LayoutLMv3 • Hugging Face • Transformers • OCR • NLP • Computer Vision • Document AI
Project Goal
The goal of the project is to move invoice processing from manual data entry toward automated intelligent document processing, while creating a system that can be further improved to handle different invoice formats and layouts.
This project represents my work across NLP, Computer Vision, Deep Learning, and Document AI, with a focus on building practical AI systems that can solve real-world business problems.
0
33
RidgeRecon — AI Fingerprint Reconstruction
RidgeRecon is an AI-powered fingerprint reconstruction project designed to restore and enhance damaged, incomplete, or low-quality fingerprint images.
The system uses deep learning and image processing techniques to reconstruct missing fingerprint ridge patterns while preserving important structural details. The goal is to transform degraded fingerprint inputs into clearer, more usable representations for further analysis.
Key Features
🧠 Deep learning-based fingerprint reconstruction
🔬 Restoration of damaged and incomplete ridge patterns
🖼️ Image preprocessing and enhancement
📐 Ridge structure preservation
⚡ Automated reconstruction pipeline
📊 Support for large-scale fingerprint datasets
Tech Stack
Python • TensorFlow/Keras • OpenCV • NumPy • Pandas • Deep Learning • Image Processing
RidgeRecon combines computer vision and deep learning to tackle a challenging image restoration problem and demonstrates practical applications of AI in biometric image processing.
#AI #MachineLearning #DeepLearning #ComputerVision #FingerprintRecognition #ImageProcessing #TensorFlow #Python #Biometrics #ArtificialIntelligence
1
181
Aegis — AI-Powered Drone Detection
Aegis is a computer vision system built with YOLO and deep learning to automatically detect drones in images and video streams. The system identifies drones in real time and provides precise bounding boxes with confidence scores for each detection.
The project focuses on developing a fast and reliable object detection pipeline that can be integrated into security monitoring, surveillance, restricted-area protection, and automated aerial monitoring systems.
Key Features
Real-time drone detection using YOLO
Accurate bounding-box localization
Confidence score for every detection
Image and video inference
Optimized computer vision pipeline
Scalable for real-time surveillance applications
Tech Stack
Python · YOLO · PyTorch · Ultralytics · OpenCV · Computer Vision
Aegis demonstrates how modern deep learning and object detection can be applied to build practical AI-powered surveillance solutions.
#ArtificialIntelligence #MachineLearning #ComputerVision #DeepLearning #YOLO #ObjectDetection #Python #PyTorch #OpenCV #Ultralytics #DroneDetection #AISurveillance #RealTimeDetection #ImageProcessing #VideoAnalytics
1
196
AutoML Forge is an end-to-end AutoML platform designed to simplify the machine learning workflow by automating key stages from data preparation to model evaluation.
The platform allows users to upload their datasets and build machine learning models without manually handling every step of the traditional ML pipeline.
Key Features
Automated data preprocessing and cleaning
Automatic feature handling and transformation
Classification and regression workflows
Anomaly detection
Model comparison and evaluation
Automated model selection
Hyperparameter optimization
Performance metrics and visualizations
Model prediction and inference
Easy-to-use interactive interface
Tech Stack
Python · PyCaret · FastAI · Scikit-learn · Pandas · NumPy · Streamlit
The project focuses on making machine learning more accessible while reducing repetitive experimentation and development time. It demonstrates practical experience in machine learning automation, model evaluation, data preprocessing, and building AI-powered applications.
Role: AI/ML Developer
Project Type: Machine Learning / AutoML / Data Science
2
1
504