Data Labeling at 24x7offshoring
, , , ,

How much data is needed for the best machine learning?

AI and device getting to know at Capital One

Data. Leveraging standardized cloud systems for data management, model development, and operationalization, we use AI and ML to look out for our clients’ financial properly-being, help them emerge as greater financially empowered, and higher control their spending.

BE equipped for AI built for enterprise data.

There’s quite a few speak about what AI can do. however what can it honestly do on your enterprise? 24x7offshoring business AI gives you all of the AI gear you want and nothing you don’t. And it’s educated in your information so that you realize it’s reliable. revolutionary generation that delivers actual-global consequences. That’s 24x7offshoring business AI.

commercial enterprise AI from 24x7offshoring


Make agile decisions, free up precious insights, and automate responsibilities with AI designed along with your enterprise context in mind.


Use AI that is skilled in your enterprise and employer information, driven via 24x7offshoring manner information, and available in the answers you use every day.


Run accountable AI constructed on leading ethics and information privateness standards even as retaining full governance and lifecycle control throughout your complete organization.

Product advantages

24x7offshoring gives the broadest and deepest set of device learning offerings and helping cloud infrastructure, putting system mastering inside the arms of each developer, statistics scientist and expert practitioner.



textual content-to-Speech
flip textual content into real looking speech.

add speech to textual content capabilities to packages.

machine learning
build, teach, and install system learning models speedy.

Translate textual content the usage of a neural machine translation carrier.

Why 24x7offshoring for AI solutions and services?

groups global are considering how artificial intelligence can help them obtain and enhance commercial enterprise effects. Many executives and IT leaders trust that AI will extensively transform their enterprise inside the subsequent 3 years — however to fulfill the desires of the next day, you ought to put together your infrastructure nowadays. 24x7offshoring main partnerships and expertise will let you enforce AI answers to do simply that.


Ai data collection
Ai data collection
Generative AI

implementing generative AI solutions calls for careful attention of ethical and privacy implications. but, whilst used responsibly, those technologies have the capacity to significantly beautify productiveness and decrease expenses across a wide variety of packages.

advanced Computing

advanced computing is fundamental to the improvement, education and deployment of AI systems. It gives the computational electricity required to address the complexity and scale of current AI programs and permit advancements in studies, actual-global programs, and the evolution and cost of AI.

Chatbots and large Language models
The talents of chatbots and large language fashions are remodeling the manner corporations function — improving efficiency, enhancing consumer reports and establishing new possibilities throughout diverse sectors.

contact center Modernization
Modernize your touch facilities by introducing automation, improving performance, enhancing consumer interactions and imparting valuable insights for non-stop development. This no longer most effective advantages organizations by way of growing operational efficiency but additionally leads to greater pleasing and personalized virtual experiences for customers.

Predictive Analytics
Predictive analytics supports groups through permitting them to make more accurate choices, reduce dangers, enhance patron stories, optimize operations and gain higher financial results. It has a wide range of packages across industries and is a treasured tool for gaining a aggressive edge in today’s facts-pushed commercial enterprise environment.

information Readiness / Governance
data readiness is vital for the a success deployment of AI in an corporation. It now not simplest improves the overall performance and accuracy of AI models however additionally addresses moral issues, regulatory necessities and operational performance, contributing to the overall success and popularity of AI programs in commercial enterprise settings.

How a good deal records is needed For machine gaining knowledge of?
records is the lifeblood of system mastering. without records, there might be no way to educate and compare 24x7offshoring models. however how an awful lot information do you need for gadget mastering? in this weblog submit, we’re going to discover the factors that have an effect on the amount of information required for an ML assignment, techniques to reduce the quantity of information needed, and guidelines that will help you get started with smaller datasets.

device gaining knowledge of (ML) and predictive analytics are two of the most important disciplines in modern computing. 24x7offshoring is a subset of synthetic intelligence (AI) that focuses on constructing fashions that may study from records rather than relying on specific programming instructions. however, statistics technological know-how is an interdisciplinary field that uses medical strategies, approaches, algorithms, and systems to extract information and insights from structured and unstructured records.

How a great deal facts is needed For machine studying?




picture by using the author: How plenty data is wanted For machine getting to know?
As 24x7offshoring and information science have turn out to be increasingly famous, one of the maximum usually asked questions is: how an awful lot statistics do you want to construct a system mastering version?

the solution to this query depends on numerous elements, together with the

  • kind of problem being solved,
  • the complexity of the version,
  • accuracy of the records,

and availability of categorised facts.
A rule-of-thumb approach indicates that it is best first of all round ten instances extra samples than the variety of capabilities for your dataset.

additionally, statistical strategies together with strength evaluation can help you estimate pattern size for diverse forms of machine-studying problems. aside from accumulating extra information, there are precise strategies to lessen the quantity of statistics wished for an 24x7offshoringversion. these encompass function selection techniques inclusive of 24x7offshoring regression or foremost element analysis (24x7offshoring). Dimensionality discount strategies like autoencoders, manifold learning algorithms, and artificial facts technology strategies like generative adversarial networks (GANs) also are available.

Even though these techniques can assist lessen the amount of information needed for an ML version, it is vital to take into account that exceptional nevertheless matters extra than amount in terms of education a successful model.

How a lot records is wanted?
factors that influence the quantity of records needed
on the subject of developing an powerful gadget learning version, getting access to the proper amount and first-rate of statistics is essential. regrettably, now not all datasets are created identical, and a few might also require extra statistics than others to broaden a successful version. we’ll explore the various factors that have an effect on the quantity of facts wished for gadget learning in addition to strategies to lessen the quantity required.

sort of trouble Being Solved
The kind of problem being solved by means of a machine getting to know model is one of the most important factors influencing the quantity of statistics needed.

as an example, supervised mastering fashions, which require categorised training statistics, will usually need greater statistics than unsupervised models, which do now not use labels.

moreover, positive kinds of troubles, which includes picture reputation or natural language processing (NLP), require large datasets because of their complexity.

The complexity of the version
any other factor influencing the amount of records needed for machine mastering is the complexity of the version itself. The more complex a model is, the greater facts it will require to characteristic successfully and accurately make predictions or classifications. models with many layers or nodes will need extra training records than people with fewer layers or nodes. additionally, fashions that use a couple of algorithms, along with ensemble strategies, will require greater information than people who use handiest a unmarried set of rules.

exceptional and Accuracy of the facts
The first-rate and accuracy of the dataset can also effect how tons statistics is wanted for gadget getting to know. suppose there is lots of noise or wrong information inside the dataset. in that case, it may be vital to increase the dataset size to get correct effects from a device-studying version.

additionally, suppose there are lacking values or outliers in the dataset. in that case, these ought to be either eliminated or imputed for a model to work successfully; thus, growing the dataset length is likewise important.

Estimating the quantity of statistics wanted
Estimating the amount of statistics wished for system studying  fashions is important in any statistics technological know-how venture. accurately determining the minimum dataset size required gives records scientists a better knowledge in their ML task’s scope, timeline, and feasibility.

when figuring out the volume of data necessary for an  version, elements along with the type of trouble being solved, the complexity of the version, the high-quality and accuracy of the information, and the provision of categorized records all come into play.

Estimating the quantity of information wished may be approached in ways:

A rule-of-thumb approach
or statistical strategies
to estimate sample length.

Rule-of-thumb approach
the rule of thumb-of-thumb technique is maximum usually used with smaller datasets. It includes taking a guess based on beyond reviews and modern expertise. but, it’s miles important to use statistical strategies to estimate sample length with larger datasets. these techniques allow facts scientists to calculate the variety of samples required to make certain sufficient accuracy and reliability in their fashions.

normally speakme, the guideline of thumb regarding machine gaining knowledge of is which you want at the least ten times as many rows (records points) as there are capabilities (columns) to your dataset.

which means if your dataset has 10 columns (i.e., functions), you ought to have as a minimum a hundred rows for premier outcomes.

latest surveys show that around eighty% of a success ML tasks use datasets with greater than 1 million statistics for education functions, with maximum utilising far greater data than this minimum threshold.

ebook a personal demo

book a Demo
information volume & high-quality
whilst determining how lots facts is needed for machine getting to know models or algorithms, you need to consider each the volume and great of the records required.

in addition to assembly the ratio noted above between the number of rows and the quantity of functions, it’s also essential to make certain adequate insurance throughout unique instructions or categories within a given dataset, otherwise called elegance imbalance or sampling bias issues. ensuring a proper amount and fine of suitable education information will help lessen such problems and permit prediction fashions trained in this larger set to gain higher accuracy ratings over the years with out extra tuning/refinement efforts later down the line.

Rule-of-thumb approximately the wide variety of rows in comparison to the wide variety of features helps access-degree information Scientists determine how an awful lot facts they ought to acquire for his or her 24x7offshoring initiatives.

thus ensuring that sufficient input exists whilst implementing system studying techniques can cross a long manner closer to keeping off not unusual pitfalls like pattern bias & underfitting during put up-deployment stages. it’s also assisting reap predictive skills quicker & within shorter improvement cycles, no matter whether one has access to significant volumes of information.

techniques to reduce the amount of records wanted
happily, numerous techniques can lessen the amount of information wished for an 24x7offshoring model. function choice strategies together with essential issue analysis (PCA) and recursive characteristic elimination (RFE) may be used to pick out and cast off redundant features from a dataset.

Dimensionality reduction techniques consisting of singular value decomposition  and t-dispensed stochastic neighbor embedding  can be used to reduce the quantity of dimensions in a dataset whilst preserving important information.

subsequently, artificial data generation techniques including generative antagonistic networks can be used to generate extra training examples from present datasets.

pointers to lessen the amounts of facts wanted for an 24x7offshoring version
further to using characteristic choice, dimensionality reduction, and artificial statistics era strategies, several different tips can assist entry-degree statistics scientists lessen the quantity of statistics wished for their 24x7offshoring models.

First, they should use pre-educated fashions on every occasion feasible because these models require less education records than custom models built from scratch. second, they should consider the use of transfer studying techniques which permit them to leverage information won from one assignment when fixing another related assignment with fewer education examples.

sooner or later, they have to try special hyperparameter settings considering some settings can also require fewer schooling examples than others.

do not leave out the AI Revolution

From facts to Predictions, Insights and selections in hours.

No-code predictive analytics for regular commercial enterprise users.

try it without spending a dime
Examples of a success tasks with Smaller Datasets
information is an critical issue of any device mastering undertaking, and the quantity of information wished can vary relying at the complexity of the model and the hassle being solved.

but, it is possible to reap a hit outcomes with smaller datasets.

we can now discover a few examples of a success projects finished the usage of smaller datasets. recent surveys have proven that many records scientists can entire a hit initiatives with smaller datasets.

according to a survey conducted by way of Kaggle in 2020, almost 70% of respondents stated they had finished a assignment with fewer than 10,000 samples. additionally, over half of the respondents said that they had finished a project with fewer than five,000 samples.

numerous examples of a hit tasks were completed the usage of smaller datasets. as an example, a team at Stanford college used a dataset of simplest 1,000 pics to create an AI machine that might correctly diagnose pores and skin cancer.

another crew at 24x7offshoring used a dataset of simplest 500 snap shots to create an AI device that might stumble on diabetic retinopathy in eye scans.

those are just examples of the way powerful machine learning fashions can be created using small datasets.

it’s miles certainly feasible to attain successful consequences with smaller datasets for gadget getting to know initiatives.

via utilising function selection techniques and dimensionality reduction strategies, it’s far viable to lessen the quantity of statistics wished for an 24x7offshoring version whilst nevertheless achieving correct outcomes.

See Our solution in movement: Watch our co-founder gift a stay demo of our predictive lead scoring tool in motion. Get a real-time understanding of ways our answer can revolutionize your lead prioritization method.

liberate valuable Insights: Delve deeper into the arena of predictive lead scoring with our comprehensive whitepaper. find out the energy and capability of this sport-changing device in your business. download Whitepaper.

experience it your self: See the electricity of predictive modeling first-hand with a live demo. discover the features, enjoy the user-pleasant interface, and see just how transformative our predictive lead scoring model may be for your enterprise. try stay .

on the quit of the day, the amount of records wished for a machine getting to know assignment relies upon on several factors, such as the type of problem being solved, the complexity of the version, the pleasant and accuracy of the facts, and the availability of labeled records. To get an correct estimate of the way a lot records is needed for a given venture, you ought to use either a rule-of-thumb or statistical techniques to calculate pattern sizes. additionally, there are effective techniques to lessen the want for large datasets, consisting of characteristic selection strategies, dimensionality discount techniques, and synthetic records technology strategies.

in the end, a success initiatives with smaller datasets are viable with the right method and to be had technologies.

24x7offshoring observe can help businesses test effects fast in gadget gaining knowledge of. it’s far a powerful platform that utilizes complete information analysis and predictive analytics to help businesses quickly pick out correlations and insights inside datasets. 24x7offshoring notice offers rich visualization tools for evaluating the satisfactory of datasets and models, in addition to clean-to-use computerized modeling capabilities.

With its person-friendly interface, corporations can accelerate the process from exploration to deployment even with restricted technical understanding. This helps them make quicker selections while lowering their costs related to growing system learning packages.

Get Predictive Analytics Powers without a statistics science team

24x7offshoring note robotically transforms your information into predictions and subsequent-high-quality-step techniques, with out coding.


  • Device mastering sales Forecast
  • popular programs of device learning in enterprise
  • A complete guide to purchaser Lifetime cost Optimization the use of Predictive Analytics
  • Predictive Analytics in advertising: everything You must know
  • Revolutionize SaaS revenue Forecasting: release the secrets to Skyrocketing achievement
  • Empower Your BI groups: No-Code Predictive Analytics for records Analysts
  • correctly Generate greater Leads with Predictive Analytics and marketing Automation

you can explore all 24x7offshoring models here. This page can be helpful if you are inquisitive about exclusive system learning use instances. sense loose to strive totally free and train your gadget learning version on any dataset with out writing code.

if you ask any data scientist how much facts is wanted for gadget studying, you’ll maximum probably get both “It depends” or “The extra, the higher.” And the aspect is, both solutions are correct.

It honestly depends on the kind of assignment you’re working on, and it’s constantly a brilliant concept to have as many applicable and dependable examples inside the datasets as you could get to get hold of correct outcomes. but the query remains: how an awful lot is sufficient? And if there isn’t sufficient statistics, how will you address its lack?

The revel in with diverse projects that worried synthetic intelligence (AI) and machine studying (ML), allowed us at Postindustria to come up with the most top of the line approaches to technique the statistics quantity difficulty. this is what we’ll communicate approximately inside the study underneath.

The complexity of a version

honestly placed, it’s the quantity of parameters that the algorithm need to learn. The extra capabilities, size, and variability of the expected output it have to keep in mind, the greater records you need to enter. as an instance, you need to train the model to predict housing expenses. you are given a desk where every row is a residence, and columns are the place, the neighborhood, the variety of bedrooms, flooring, bathrooms, etc., and the fee. In this example, you educate the version to predict fees based on the trade of variables in the columns. And to learn how each additional input characteristic affects the input, you’ll want greater facts examples.

The complexity of the mastering set of rules
greater complicated algorithms always require a larger amount of records. in case your undertaking wishes widespread  algorithms that use based mastering, a smaller quantity of statistics could be sufficient. Even if you feed the algorithm with greater statistics than it’s enough, the results received’t enhance notably.

The scenario is one of a kind with regards to deep mastering algorithms. unlike conventional system gaining knowledge of, deep gaining knowledge of doesn’t require function engineering (i.e., building enter values for the model to match into) and is still able to examine the illustration from raw information. They work without a predefined shape and determine out all of the parameters themselves. In this case, you’ll want greater records that is relevant for the algorithm-generated classes.

Labeling desires
depending on how many labels the algorithms ought to are expecting, you could need numerous amounts of enter facts. as an example, in case you want to type out the pix of cats from the photographs of the puppies, the algorithm desires to learn some representations internally, and to do so, it converts enter facts into these representations. however if it’s just locating pics of squares and triangles, the representations that the algorithm has to examine are easier, so the amount of statistics it’ll require is much smaller.

suitable errors margin
The type of undertaking you’re operating on is another thing that impacts the quantity of records you need due to the fact one of a kind tasks have extraordinary levels of tolerance for mistakes. as an example, if your venture is to are expecting the weather, the algorithm prediction can be misguided by some 10 or 20%. however when the set of rules ought to inform whether or not the patient has most cancers or no longer, the degree of blunders may cost a little the affected person lifestyles. so you need more data to get more correct outcomes.

input range
In some instances, algorithms need to be taught to characteristic in unpredictable conditions. for instance, when you broaden an online virtual assistant, you evidently need it to recognize what a traveler of a company’s internet site asks. but humans don’t generally write flawlessly correct sentences with standard requests. they may ask hundreds of different questions, use special patterns, make grammar mistakes, and so on. The more out of control the environment is, the greater information you want on your ML undertaking.

based at the elements above, you may outline the scale of information sets you need to acquire properly set of rules overall performance and dependable results. Now allow’s dive deeper and find a solution to our predominant question: how much data is required for gadget gaining knowledge of?

what’s the most beneficial size of AI schooling information sets?
whilst making plans an ML assignment, many fear that they don’t have quite a few statistics, and the outcomes gained’t be as dependable as they can be. however only some sincerely recognise how a lot facts is “too little,” “too much,” or “sufficient.”

The maximum commonplace manner to outline whether a statistics set is sufficient is to apply a 10 instances rule. This rule method that the quantity of enter information (i.e., the wide variety of examples) must be ten instances extra than the quantity of stages of freedom a version has. usually, stages of freedom imply parameters for your statistics set.

So, for example, if your algorithm distinguishes pix of cats from snap shots of dogs based on 1,000 parameters, you need 10,000 pictures to teach the version.

even though the ten times rule in device gaining knowledge of is pretty popular, it can best work for small fashions. large models do no longer observe this rule, as the range of amassed examples doesn’t always reflect the actual amount of schooling statistics. In our case, we’ll want to matter now not most effective the range of rows however the variety of columns, too. The right approach would be to multiply the wide variety of photographs by way of the size of every picture with the aid of the quantity of colour channels.

you could use it for rough estimation to get the assignment off the ground. however to discern out how much facts is needed to educate a specific model inside your particular undertaking, you need to find a technical companion with applicable know-how and visit them.

On top of that, you always have to remember that the AI models don’t observe the records but as a substitute the relationships and patterns in the back of the statistics. So it’s now not only amount in order to have an impact on the results, but also high-quality.

however what are you able to do if the datasets are scarce? There are a few strategies to cope with this trouble.

a way to cope with the dearth of statistics
loss of facts makes it not possible to set up the family members among the input and output records, for that reason inflicting what’s called “‘underfitting”. if you lack input statistics, you could either create synthetic statistics units, increase the existing ones, or observe the information and records generated earlier to a similar hassle. allow’s overview every case in greater element beneath.

records augmentation
facts augmentation is a method of increasing an input dataset by means of slightly converting the prevailing (authentic) examples. It’s extensively used for picture segmentation and category. typical picture alteration strategies consist of cropping, rotation, zooming, flipping, and color modifications.

How a great deal records is required for device studying?
In wellknown, records augmentation facilitates in fixing the hassle of restrained statistics by way of scaling the available datasets. except image classification, it could be utilized in a number of other instances. for instance, right here’s how statistics augmentation works in natural language processing :

back translation: translating the textual content from the authentic language into a goal one after which from target one lower back to authentic
clean data augmentation: changing synonyms, random insertion, random swap, random deletion, shuffle sentence orders to receive new samples and exclude the duplicates
Contextualized phrase embeddings: education the algorithm to use the phrase in distinctive contexts (e.g., while you need to apprehend whether the ‘mouse’ means an animal or a device)

information augmentation adds greater flexible records to the fashions, helps remedy elegance imbalance troubles, and increases generalization potential. but, if the original dataset is biased, so could be the augmented records.

synthetic records generation
synthetic records technology in machine mastering is every so often considered a sort of records augmentation, however these concepts are different. throughout augmentation, we alternate the characteristics of facts (i.e., blur or crop the photograph so we can have 3 images as opposed to one), even as synthetic generation manner creating new facts with alike but no longer similar homes (i.e., growing new snap shots of cats based at the preceding snap shots of cats).

at some stage in artificial information era, you may label the information right away and then generate it from the supply, predicting precisely the records you’ll receive, that’s useful whilst no longer a good deal information is available. but, at the same time as working with the real statistics units, you want to first acquire the facts and then label every instance. This synthetic statistics era technique is widely applied when developing AI-based totally healthcare and fintech answers when you consider that actual-existence data in these industries is challenge to strict privateness legal guidelines.

At Postindustria, we also observe a synthetic information method

Our current virtual jewelry strive-on is a top example of it. To broaden a hand-monitoring model that could work for diverse hand sizes, we’d want to get a pattern of fifty,000-a hundred,000 arms. when you consider that it might be unrealistic to get and label such some of actual snap shots, we created them synthetically by way of drawing the pictures of different arms in numerous positions in a unique visualization program. This gave us the vital datasets for schooling the set of rules to song the hand and make the ring suit the width of the finger.

whilst artificial records can be a great answer for lots projects, it has its flaws.

synthetic statistics vs real facts problem

one of the problems with synthetic information is that it is able to lead to results which have little software in fixing actual-existence problems when real-existence variables are stepping in. for instance, in case you increase a virtual makeup attempt-on using the pics of humans with one pores and skin colour after which generate more synthetic data based on the existing samples, then the app wouldn’t work well on other skin colours. The result? The customers won’t be happy with the characteristic, so the app will reduce the range of capacity shoppers rather than growing it.

some other difficulty of having predominantly synthetic information deals with producing biased effects. the bias may be inherited from the unique sample or when different factors are overlooked. as an instance, if we take ten people with a certain fitness circumstance and create greater information based on the ones instances to expect what number of human beings can increase the identical circumstance out of 1,000, the generated facts might be biased due to the fact the authentic sample is biased by the selection of range (ten).

transfer studying

transfer learning is any other approach of solving the hassle of restrained data. This approach is based totally on applying the knowledge received when operating on one challenge to a new similar venture. The idea of transfer gaining knowledge of is that you teach a neural network on a particular facts set and then use the lower ‘frozen’ layers as characteristic extractors. Then, pinnacle layers are used train different, more specific statistics units.

For example, the version changed into skilled to apprehend photographs of wild animals (e.g., lions, giraffes, bears, elephants, tigers). subsequent, it can extract capabilities from the similarly snap shots to do greater speicifc evaluation and understand animal species (i.e., may be used to distinguish the snap shots of lions and tigers).

How a great deal records is needed for machine learning?

The switch getting to know technique quickens the education degree because it permits you to apply the spine community output as functions in in addition levels. but it can be used simplest while the tasks are comparable; otherwise, this approach can have an effect on the effectiveness of the version.

but, the provision of information itself is frequently not enough to correctly educate an  version for a medtech answer. The fine of records is of maximum significance in healthcare initiatives. Heterogeneous information sorts is a assignment to investigate in this discipline. statistics from laboratory assessments, medical photos, vital symptoms, genomics all are available in one of a kind formats, making it hard to installation ML algorithms to all of the information straight away.

another trouble is wide-unfold accessibility of medical datasets. 24x7offshoring, for instance, which is taken into consideration to be one of the pioneers inside the area, claims to have the most effective notably sized database of vital care health data that is publicly available. Its 24x7offshoring database stores and analyzes health information from over forty,000 essential care patients. The information include demographics, laboratory exams, vital symptoms accumulated via patient-worn video display units (blood pressure, oxygen saturation, coronary heart rate), medications, imaging facts and notes written via clinicians. another strong dataset is Truven fitness Analytics database, which records from 230 million patients collected over 40 years based totally on coverage claims. but, it’s not publicly available.

every other problem is small numbers of statistics for some sicknesses. figuring out disorder subtypes with AI calls for a enough amount of facts for each subtype to teach ML fashions. In some instances data are too scarce to train an algorithm. In those cases, scientists try to increase ML fashions that learn as plenty as possible from healthful patient statistics. We must use care, but, to make sure we don’t bias algorithms toward healthy patients.

need facts for an24x7offshoring mission? we are able to get you blanketed!
the size of AI education data sets is crucial for gadget gaining knowledge of initiatives. To outline the most reliable amount of information you need, you have to consider loads of things, inclusive of mission kind, algorithm and model complexity, blunders margin, and enter range. you can also follow a ten instances rule, but it’s now not constantly dependable in relation to complicated responsibilities.

in case you finish that the available facts isn’t sufficient and it’s not possible or too high priced to collect the required actual-world statistics, try to apply one of the scaling techniques. it can be facts augmentation, artificial facts generation, or transfer studying relying on your project desires and finances.

some thing choice you pick out, it’ll want the supervision of experienced facts scientists; otherwise, you risk finishing up with biased relationships among the input and output information. this is where we, at 24x7offshoring, can assist. contact us, and permit’s communicate approximately your 24x7offshoring project!

Table of Contents