top of page

DeepNotes is a Platform as a Service (PaaS) for curating knowledge from multiple data sources to help businesses make optimal and accountable assessments and decisions. It comprises of a comprehensive technical stack of NLP, ML, graph database, cloud computing technologies as well as a set of pre-trained models.

We are proud of our strong R&D background and the technologies we built!

Scanning contents from multiple data sources at scale


Scanning contents at scale

  • Leveraging big data technologies to scan thousands of documents at scale

  • Connecting to multiple data sources in the cloud and on premise

  • Periodically crawling public websites to retrieve and scan public datasets


Parsing a large variety of file formats

  • Parsing most of common file formats to extract textual content and metadata without any configuration

  • Supporting all Microsoft Office documents, PDFs, HTMLs, CSV, image formats.

  • Plugging in custom components to extract specific content parts from documents 


Discovering information from rich contents


Text cognition

  • Classifying documents and extracting named entities from texts

  • Matching semantically smilar texts and similar cases across documents

  • Finding factoid answers from piles of documents

  • Summarising texts

  • Leveraging BERT language model and its variants through HuggingFace libraries

  • Supporting zero-shot training NLP

  • Leveraging rule-based NLP framework

  • and much more ... ...


Form cognition

  • Discovering content object by its textual meaning and geometric layout in page

  • Integrating with OCR to transform image to text

  • Applying NLP to extract and index form objects

  • Detecting forms and tables

  • Reacting and marking form content objects

  • Validating fields in forms

  • and much more ... ...


Table cognition

  • Using ML to classify database tables, columns, rows and fields

  • Applying NLP to scan and understand textual fields in tables

  • Modelling business intents of tables

  • Detecting table similarities

  • Scaling up metadata tagging for enterprise data lakes


Face cognition

  • Detecting faces in images and videos

  • Matching faces

  • Detecting facial expressions in videos (e.g. smiling, blinking eyes)

  • Detecting facial emotions in image and videos (e.g. sad, happy)

  • Integrated with mobile phone cameras

  • Reacting faces in videos and images


Image cognition

  • Matching images and de-duplicating images in image collections

  • Detecting and marking image differences

  • Assessing image qualities

  • Leveraging large pre-trained image models

  • Classifying images with support of training custom classifier

  • Detecting objects in images with support of training customer image detector


Voice cognition

  • Transcribing voice to text

  • Detecting speaker using voice signature

  • Applying NLP to transcribed text

  • Indexing speaker with spoken content


Transforming information to knowledge


Linking disparate information as knowledge graphs

  • Transforming discovered information to uniformed knowledge graphs

  • Linking up information captured in different digital content from disparate sources

  • Injecting annotations to orginating content sources to trace knowledge to data


Encoding rich contextual data in linked information

  • Encoding rich contextual data in graph nodes 

  • Using word embeddings and sentence embeddings to represent meanings of a text chunk

  • Enriching knowledge graphs with related policy and guideline information

  • Enriching knowledge graphs with historical data

  • Supporting accountability through linked and enriched information


Reasoning about curated knowledge and advanced analytics


Querying & visualizing knowledge graphs

  • Using powerful graph query language to query knowledge graphs

  • Drilling down knowledge graphs to their originating data

  • Visualizing knowledge graphs in either 2D or 3D


Recommending and suggesting similarities

  • Efficiently matching similar cases based on knowledge graphs

  • Recommending items by matching knowledge graphs and embeddings

  • Supporting both rule-based and GNN-based recommendation


Detecting anomalies

  • Detecting anomalies in knowledge graphs

  • Training GNN model based on knowledge graphs

  • Leveraging embeddings attached to nodes in anomaly detection and recommendation


Fasttracking Implementation and Lowering Technical Barriers and Risks


Cloud and on-premise deployment with Devops

  • Deployed natively to AWS EKS

  • Deployable to on-premise environment or any cloud supporting Kubernetes

  • New environments provisioned in minutes with preconfigured templates

  • Scripted deployment and updates of applications and models


Pretrained models

  • Using multiple open source pretrained libraries

  • Extensive use of HuggingFace libraries for NLP tasks

  • Seamlessly integrating AI services available on AWS

  • Proprietary trained models available on demand


Unsupervised learning and augmented AI

  • Training and utilizing custom models through Tensorflow

  • Supporting zero-shot and few-shot learning

  • Utilizing unsupervised learning and transferred learning to lower barriers of preparing training data

  • Using rule-based NLP tools to automate preparation of training data

bottom of page