Digital transformation with AI and process automation.
Prior consulting use cases in the domain of talent acquisition, e-commerce, e-Publishing and HR analytics.
1. Digital Transformation
with AI and Process Automation
Harendra Singh
AI & AUTOMATION CONSULTANT
harendra@lueinanalytics.com
LUEIN ANALYTICS
2. About
A thought leader and machine learning enthusiast with great passion for technology, creativity, challenging goals &
social entrepreneurship. Consulted more than 20+ startups in the areas of Natural Language Processing, Text/Image/
Video/Audio Processing, Deep Learning, Data Mining, Machine Learning, Unstructured Document Parsing, Statistics
or related field
Core expertise in :
1. AI and Automation product development
2. AI team hiring and team building for clients
3. AI training and team skill development
4. Remote AI product development and consulting
5. Generating new revenue stream for client and also helping client to save a lot of cost with
automated solutions.
3. Optimized the logistic packing algorithm
Select the best and most efficient box for a shipment
● Dependent factors - shipper types, dimensions of items,
rotations, packing more than one items at a time, shipping
costs, operation time, and experience for both our clients
and their customers.
● Ensemble model with combination of below algorithms:
○ First Fit Descending, pack the biggest products
first in the smallest space we can.
○ Knapsack problem, given a set of items, each with
a weight and a value, determine the number of each
item to include in a collection so that the total
weight is less than or equal to a given limit and the
total value is as large as possible.
○ Bin packing problem, objects of different volumes
must be packed into a finite number of bins or
containers each of volume V in a way that minimizes
the number of bins used.
4. Auto sectioned sample PDF output
PDF content extraction starts with manually zoning all the sections
( like title, images, sub-titles, page number, paragraphs, etc ) in a
PDF. Then it goes for section by section content extraction
With image processing and convolution neural network, we have
automated the entire manual zoning process.
We treated the text problem into an image problem and solved
the manual zoning problem effectively.
Auto PDF section identification and extraction
5. AI for video hiring ( AUDIO / SPEECH / VIDEO / IMAGE )
AI for video interview and video resume from saved or
real time interview streaming
● Video content parser
● Video content analytics
● Calculate candidates confidence level
● Calculate discussion interest level
● Calculate subject understanding level
● Calculate communications levels
● Calculate solution approach strategy
● Calculate calculate response time to answer
● Check whether candidate copying during interview.
● Candidate nervousness (with facial micro expressions)
● Conduct a expert less interview process.
6. Text summarization and generation (under process)
We are currently working on sophisticated Neural
Network algorithm combined with supporting framework
and infrastructure, we running multiple researches to
solve below Publishing use cases in the space of Natural
Language Generation:
1. Automatically create an initial article for author
from a topic
Create unique content from scratch, simulating a
human writer. You choose the topic and length, and the
algorithm will create your textual content.
2. NLG Financial Services
Automatically generate high quality personalised risk
analysis, financial, compliance and other reports in
writing in seconds..
3. Academic books summarization
Automatically generate titles, rewrite articles and
content summarisation for academic books.
7. Content generation
Machine scans an image and generate an alternate
description for the same.
● Significant amount of scientific research has been
carried out and still under process, to solve one of the
toughest problem of image alternate text generation
(also called as image caption generation).
● Machine takes an images as an input and returns a
short or long description about the input image.
● Neural network based algorithm image classification,
object detection, computer vision processing, along with
natural language generation has been used to solve this
problem .
● Image captioning domain is quite vast, we have
achieved initial success in medical equipment images,
maths graphs/reports and complex object detection.
8. 1. Extract faces and names from school yearbook scanned image
dataset (collected from US, UK and few more European countries
school year books since past 45 years) and create facial database for
the client automatically.
2. Detect face from individual face images and also from group
scanned yearbook image files.
3. Extract Names and alias names from the same scanned yearbook
image file.
4. Using natural language processing and custom business rule to
associate faces with their names and feed it into the database.
Face detection, extraction and auto tagging
Sample input image file
9. Tone and actor
labeled srt file
Extract dialogue
and create subtitle
file
Extract
speech
features
Translate srt file
Process audio
for tone and
features
Embed tone feature in
translated audio
Speech trained
model
Srt file
Input Video
Video auto subtitleing.
Video emotion extraction (emotionML).
Audio feature extraction.
Tone analysis.
Subtitling translation to another language.
Use text2speech library to convert translated text to another language speech/dialogue text.
Embed original tone intensity features from original audio to new audio, also embed video emotion features
to make the final audio output look more realistic.
Embed new audio to original video file to generate dubbed video file.
Use minor manual modulation (if needed) to sync the final dialogue speed and timings
Video auto-dubbing (under development)
Use tone features
Embed new
audio to video
with dialog time
Dubbed Video
Tone analysis injection
text2speech
10. Generate alternate text for an image
Neural Network based generative model for captioning/
describing an image automatically in natural language.
Machine learning model trained with supervised 20k plus
manually labeled images.
The process consist of three core components :
1. CNN encoder
A pre-trained CNN is used to encode an image to its
features and also pre-encoded each image to its feature set
for high performance and speed.
2. Word embedding
Used pre-trained word embedding model and also explicitly
trained and embedding model that takes a word and
outputs an embedding vector.
3. CNN decoder
It takes the image vector and partial captions at the current
timestep and input and generated the next most probable
word as output.
11. Research and work on multimedia ( AUDIO / SPEECH / VIDEO / IMAGE )
1. VIDEO PROCESSING
Automatic subtitling for TV series videos.
Video search within for tagging opening credits and closing
credits.
Video dubbing- translating and embedding speech features in new
audio (under research).
Real time emotion and expression identification for complex
behavior identification.
2. SPEECH PROCESSING
Audio Signal Classification, speech detection, audio tone analysis,
dialogue recognition, speech-to-text and text-to-speech
conversion.
act text with layout from scanned documents.
3. IMAGE PROCESSING WITH NEURAL NETWORK
Object detection, face detection & recognition, image similarity
check and image upsampling / downsampling
4. CONTENT EXTRACTION FROM IMAGE
Text area detection.
Automatic layout analysis.
Object detection & labelling.
Content extraction from scanned financial forms.
5. IMAGE CAPTIONING
Object detection, brand detection, event detection and
automatically generate alternate-text for an image.
6. SMART OCR
Text/image/table detection & extraction. OCR
orientation detection & correction. Extract text with
layout from scanned documents.
Worked on many multimedia AI use cases. Listing all the recent individual research and client use cases as below :
12. 1. Adaptability is more vital to success than ever.
2. Smart machines and artificial intelligence (AI) are taking off in a big way.
3. Growing importance of the user experience.
4. Innovating rapidly and AI is being used across organisations.
5. Organisations are embracing remote AI workforces.
6. AI growth is sustainable and should not Be Ignored
WHY COMPANIES NEED AI DIGITAL TRANSFORMATION