Sessions

Metorikku is an open-source library that simplifies the writing and execution of ETLs on top of Apache Spark. Using the Metorikku library, users can create complex ETL jobs by only relying on simple configuration files and SQL commands. Moreover, the library includes an infrastructure for creating unit tests. In this talk, we will show how we leveraged Metorikku at Yotpo and made spark accessible to our entire R&D.

Data science and machine learning have become very popular recently and some of the “magical black-box” methods such as deep-learning became common buzz-words in tech industry slang. But in moving from data science towards leading people, you suddenly realize that there are things that you cannot apply directly from online courses. To make this transition smoother, I propose to “transfer learn” machine learning jargon into team management. In this talk I’ll share a few machine learning concepts that can help you understand team behaviour better and explain how to deal with the “dropout” of your team members and avoid “overfitting” your approach to team management.

Hiring well in a competitive market is a hard and time consuming task. Moreover, we all want to hire someone which has the technical potential but also fits the team and culture. In this session, I’ll briefly review the main aspects of hiring engineers and drill down to three aspects which are crucial, but are often overlooked: screening for culture, building the right hiring process and how to learn and constantly improve.

Let's create a game set in space!

For this, I have to render a nice looking galaxy. How do I do it? Do I create a pre-rendered image and place 100k stars manually (oh no!), or do I try to create something random procedurally that looks awesome?

In this talk, I'll delve into the basic concepts of procedural generation - mostly around the basic mathematical tools at your disposal as a developer to create awesome procedural generated art.

With every step in creating a realistic looking galaxy, I will provide code examples in JavaScript.

When AI/ML model goes to production - it immediately starts to have its own life and meet the reality of the production-mess data streams. Without the proper monitoring - a big mess is just a matter of time, letting the algorithm take decisions on GIGO basis (Garbage-In-Garbage-Out). Looking at the traditional software/infrastructure KPIs will just dont do it. In this session, we'll share a few of the most important KPIs to monitor when you launch your AI/ML in a production environment.

“Why on earth did we decide to develop our own RPC system?” is a simple and obvious question to ask, but a complex one to answer. In this talk, Dr. Robert Fink, Head of R&D at Palantir Technologies, will give a firsthand account of the development history of Palantir's polyglot RPC system, Conjure, from the first line of code to its rebirth in open-source. We discuss the rationale and design principles behind Conjure, its implementation, and the lessons learnt on the way (“self-hosting intermediate representation”, anyone?). If you are looking for a simple yet strongly-typed RPC system that works for servers and clients in Java, Go, Rust, Python, and TypeScript, maybe look no further than Conjure?

Tracking users is not only IP recording. You can use a lot of tools to track and analyze users across platforms even without IP. In this session, I will explain how users try to conceal their identities and how to bypass those techniques. How you can detect users and track them even if the use incognito mode. How to track them even if they use VPN or TOR. There are a lot of open source tools and code that is available to help us: from digital fingerprint to forever cookies. From WebRTC local IP detection to DNS cookies. You can use the information in this session to violate's users privacy, or you can use the session information to know how to defend your privacy. I use it for the latter.

What would Node.js look like if it was written in 2019?

Deno is a new prototype project by the original creator of Node.js - Ryan Dahl.

It was created with lessons/mistakes learned from Node.js in mind. In this talk we’ll talk about the future of Deno, its philosophy and how to check it out.

Streaming algorithms are everywhere, and are relevant for any developer that has to deal with scale - whether it's message streams or queries on big data.

While many developers will just treat them as black magic, a slightly more in-depth knowledge of them can be interesting and critically useful.

In this talk, I intend to:

• Explain what streaming algorithms are,
• Show how, in Glassbox, we used them in original ways to implement otherwise impossible solutions,
• Present the difficulties and surprises,
• Present a small catalog of useful algorithms that you could find original uses for.

After this talk, you will have powerful new tools in your coding arsenal.

At Firefly.ai we use autoML to build models for users from various industries. These models can achieve high scores, but this usually comes at the price of complex models that are hard to interpret. Understanding how our models make their predictions is important for quality control, revealing patterns in the data, etc.

Interpretability of ML models is a relatively new field that is researched through a variety of aspects. In this session I will review how we applied techniques such as sensitivity analysis, PDP & ICE, and LIME for explaining the predictions of generic, black-box models at Firefly.ai. Using two real-world use cases from the banking and real-estate industries, I will describe the pros and cons of each method, differences between the methods, and production considerations.

Optimizing an application’s performance for the underlying machine is a difficult task. Increasing hardware complexity, workload diversity, and the unmanageable volume of data produced by performance tools increase the optimization challenges. At the same time resource and time constraints get tougher with recently emerged segments.

This session overviews the Top-down Microarchitecture Analysis (TMA) method and its handling of cycle accounting in modern out-of-order cores. It illustrates some performance problems that call for truly top-down-oriented metrics, presents recent challenges of modern data centers, and performance monitoring unit (PMU) enhancements to address them

During the past 8 years, Fiverr migrated from a LAMP architecture all the way to 200~ Microservices with more than a dozen teams pushing a few hundred deploys a week.

Working inside the monolith was simple, all data was available but the move to microservices presented challenges as data was exposed via REST API, which caused performance and reliability issues, dependencies between teams and much more...

If you feel that your current strategy for sharing data is brittle this talk is for you! I'll present our robust solution for sharing data across Microservices, utilizing Kafka as a data hub using techniques borrowed from CQRS and DDD to tie everything together and iterate much faster!

We are engineers.

Or at least rationale, realistic and methodological creatures. We are great at optimizing CPU cycles, load times and conversion rates -- but what about our own career path?

During our professional career, we are given countless opportunities to steer it towards its optimal course.

But what are we optimizing exactly? What are the choices we are given? What is the optimization function? What hidden factors are part of it? Are there known "right" or "wrong" choices?

Let's optimize!

"We have a problem in production! does it reproduce in staging?" is a sentence often uttered in many companies. Underlying this simple sentence lie implicit assumptions about environments which often turn out to be false in our modern services and applications. Clusters and micro-services complicate matters and make production environments hard to reproduce and model, but when A/B tests, feature toggles and ephemeral containers, serverless functions and multi-tenancy are introduced one must conclude we no longer have a "production environment", but rather "production environments", plural. In a sense, every transaction experiences a unique unreproducible universe. In this world, what is the value of Staging environments? how does such understanding change our development processes?

In 2015, compromised developers unwittingly used a malicious XCode to inject malware into thousands of apps, stealing data of millions of users. In both 2014 and 2017, millions of Uber user and driver records were leaked following an improperly secured token commited to a GitHub repo. And in 2013, a phished developer was tricked to give the Syrian Electronic Army to the Financial Times’ site.

What do all of these have in common? We, developers, caused them.

In this talk, we’ll better understand the damage we can cause, from distributing trojans to exposing user code. We’ll better understand the techniques attackers use to compromise us, so we can be prepared. Most importantly, we’ll talk about how we can defend ourselves – and our users.

Measuring unique users in a billion user network is hard - accurate counting is space consuming and not easily distributable.

In this talk I will describe HyperLogLog, a probabilistic cardinality estimation algorithm and data structure and how we used it to provide breakdowns of our billion user reach.

What does it mean to strive for ENGINEERING EXCELLENCE? As your engineering organization grows, you want to encourage teams and individuals to develop and improve their technical proficiency and practices. Growth puts stress on your teams' ability to deliver kick-ass code consistently. To help them improve and grow, we built the LEVEL UP model to represent everything that's expected of an excellent engineering team.

In this talk, I’ll review why we built this model, refined it and how we got teams to adopt it and the effects that it had on our department. I will also give you the tools to either use our model or fork your own specific one for your organization.

At JoyTunes we run a fairly unique startup-within-a-startup pod structure. Harnessing lessons learned from Spotify, Netflix and a few others, we built the entire company around independent, fast moving and self sustained pods.

In this talk we’ll use real life examples, including listing the pods we currently have, projects we’ve engaged in this manner and successes/failures. We'll briefly cover the reasoning and core principles along with the wonderful benefits and the not-quite-as-wonderful challenges, after doing this (successfully) for a little over a year.

Feature flags and user configuration systems are important! They allow PMs and developers to fully control their product behavior, run experiments, tweak user experience and handle a crisis.

When we wanted to enhance our product delivery process at Soluto, we found out that publicly available feature management platforms are in short supply - most companies invest lots of engineering power for building their own in-house solutions.

In this session, I’ll showcase and demo Tweek - our open source, rich, user-friendly feature manager that can be easily used by everyone, anywhere!

Only half of a mobile app's users use it more than once. Maybe they just got what they needed, but often they leave because they’re overwhelmed, not sure what to do or think they can’t get what they came for.

Good onboarding and introduction can drastically improve this, but it’s not easy to make it work for everyone. Some users like to be guided and hand-held, while others just want to explore. How can we support both types of beginners? Can we avoid a Clippy situation, an onboarding process that detracts more than it helps?

I spent 2 years at the MIT Media Lab investigating how to make it easier for kids to get started with Scratch, a coding environment for kids. In this talk I’ll share some key points learned along the way and demonstrate how to apply them in any onboarding process.

Predictive innovation is something humanity has always been struggling with.
 We, as product builders, lean too much on gut feeling and a bit of luck when launching new products or introducing new features.

 People don't just buy products; they hire them to perform a certain job.

 In this session, I will introduce Jobs-To-Be-Done theory that helps us understand the causality behind people's choices and deconstruct their decision-making process so we can create a replicable process to uncover people's unmet needs and build better products for them

You fear your code. Once in production, you are afraid of changing it. A bug? You fix it using messy, yet surgical, hacks, rather than a refactoring that leaves your code better.

Admit it! You fear your code. But there's a way out: just write tests. Writing tests is impossible! It makes my productivity tank! My boss doesn’t leave me any time for writing tests!

These are all excuses for not writing tests. I will show you techniques that will force you to write those tests, and I will show you how they help you sleep better at night, make you more productive, and make you a much better developer.

Machine learning and AI received a lot of attention in the last few years, but very little was published on how to architect real-life ML-based systems.

In this session we present our architecture that supports an AI-based system with numerous live models handling multiple types of data (audio, speech, text, numeric data) with multiple training and inference lifecycles, and multiple approaches (supervised vs. unsupervised, traditional vs. neural networks).

We discuss how we created a cloud architecture supporting security and privacy, how we built a system architecture that supports training and inference at scale, and how we package software in a polyglot (Java/Python/...) environment.

How, when and where did the podcast and the conference came to be. Stories behind the scenes and the unwritten (unspoken) history of technical content in Israel’s tech

A common misconception is that data science is only about choosing the right model for the problem at hand, when, in fact, considerable time and effort are put into more fundamental challenges. One major bottleneck in machine learning is getting reliable labeled data to train the model on.

How to learn on partially labeled data? How to deal with a data set that may potentially contain miss-labeled observations? How to create a feedback loop on your model predictions?

These are some of the labeling challenges that we are dealing with in PerimeterX. Many of them don't have a single, clear solution, and can be approached from different angles. Our talk will review some academic approaches that discuss these issues alongside case studies from PerimeterX.

I never imagined I’d know who Indian cricket star Rohit Sharma is, but then traffic to our real-time HTTP request sender service suddenly increased from 20 to 40 million events per minute. The reason? An app streaming the first game of the Indian cricket season. While growth is a good thing, we found ourselves unprepared for this sudden spike and we needed to scramble. Initially, we just threw money at the problem, but this wasn't sustainable. My talk will describe how we found low-cost, programmatic and architectural solutions to this problem and how we prepared ourselves to handle massive spikes like these on top of our existing 70 billion events per day. I'll explain our process of profiling, performance enhancement techniques, and some important lessons learned along the way.

In 2011 I was the PM for the team that powered the Google+ social graph (such as it was.) In this session, we will post-mortem a previously untold incident, in which my team developed a new algorithm for suggesting friends, which we had hoped would catapult the social network on a new trajectory to success. We soon found out, in the middle of the night, that we had miscalculated badly. We will learn how optimizing for the wrong metric can spell doom even if the engineering is flawless.

Early detection of abnormal events can be critical for many business applications, however there are numerous challenges when implementing real-time anomaly models at scale. Moreover, most analytical models have been traditionally designed for the batch processing paradigm and usually cannot be easily adapted to unbounded datasets and real-time latencies.

At PayPal, we've built a generic framework for developing robust and scalable anomaly detection streaming applications, focusing on flexibility. Inspired by the design of scikit-learn and Spark MLlib, we've designed a simple pipeline-based API on top of Spark Structured Streaming, to capture common patterns of the anomaly detection domain.

When I started using git 7 years ago, I was baffled by its mysterious jargon, and intimidated by the occasional French-revolution threat You are in detached HEAD. WAT?

I do like my head attached, so I kept the secret industry best practice: stick to add, commit, pull and push, and if anything goes wrong - copy the repo to another dir, pray, and apply some online black magic. This sucks. I decided to learn git from first principles.

In this session I will explain the git fundamentals: file structure, SHAs and objects, commits, refs and branches, up to the git interactive rebase. After this session, you should be able to understand git man pages, error messages, and repo states - without losing your HEAD.

As our SQL table storing the customers of our users - our users being small to medium business owners - grew too large to be performant we needed to start looking at other solutions for storing a large volume of data. But our data is not just large in quantity, that would be too simple. Our data is somewhat relational in it's nature, it changes often, it needs to be searched effectively and fast, and it needs to be GDPR compliant. In this talk you will learn about our stack, about acceptable limitations to the system as well as some unexpected caveats we encountered, while designing, implementing and shipping to production a large scale, searchable, available and near real time system.

Part of the software developer job is to find new and better ways to solve problems. Writing code using his/her wits, intelligent and creativity.

However, sometimes being too clever can lead to hard to track bugs, maintainability issues and impossible to understand code. Is all cleverly written code good code, or is it a problem just waiting to happen?

In this session, I will show you real world examples of cleverly written code. And show you how we can use clean code principles, refactoring and design patterns, to transform that code from clever code to good code, code one that your peers - and future self - would thank you for writing.

Software is eating the world. We, developers and architects, are a major force influencing software, technology, and the world it creates. We don’t have the privilege of being unaware of our actions. If we really want to create a better world, we must understand the intersection of technology and humanity. We need to open our eyes to the link between ethics and software. In this session we’ll look at some examples of ethical questions involving software and algorithms. We’ll discuss technology, sense of self, politics, truth, and try to understand what we can do about it.

Deep Learning brings with it enormous amounts of data, complicated experiment results and intense compute requirements. Decades of experience in moving code to production yielded best practices in engineering that have not yet found their place in deep learning teams. Over the past few years, the term “DevOps” has become an industry standard but where are the DeepOps best practices for AI First companies? In this talk, we’ll cover why most AI companies fail when it comes to reproducible training, data management, and deploying to production.

Building modern web apps require a lot of Javascript, which is necessary for making them intractable. Delivering JavaScript efficiently by splitting your bundles smartly, can shorten your time to intractability and overall load times significantly. In the session I will share how Javascript bundling work at Facebook, and how you can implement similar ideas in your web app.

So, you are 27 years old and already "at the top of your career." Awesome! But, what are you going to do for the next 40 years until you retire?

Oh and wait, we all know there are almost no >50 year old people in this industry, does it mean that you only have ~20 years of career left?

Your salary is great and you like going to work every day, but is staying where you are the right decision for you in the long term?

Are there any steps you can take to make sure your career is moving in the path which is right for you?

Of course!

In this talk I'm going to offer my perspective on how you can handle these decisions to get your career on the right path.

Can you expect your engineers to act as managers? Is it too high of an expectation or will it help them grow? How will it impact your guidance and ability to closely work with them?

In this talk, I will share my experience, insights and conclusions as a manager of managers on this matter.

Amazon introduced EC2 back in 2006. Since then, additional vendors introduced similar public cloud offerings. As a result, new applications and businesses were created based on the new features, cost and business models that these public clouds are offering. The question of private vs public deployments keeps being considered in many new as well as existing applications. The simple logic of "this deployment costs X in private data center and 3X in public cloud deployment" is missing some of the key consideration like: agile architecture changes, flexible consumption models and will likely lead to wrong conclusions. This talk will focus on a several considerations, based on real life experience, on when it makes sense to use the private data-center and when to use the public cloud.

Writing user input validations is something most front end developers get to deal with during their careers. For some reason, this simple enough task often adds complexity and bloat into the developer's code.

We borrowed concepts and syntax from the world of JS unit tests to create an open source framework that handles the most common pains of user input validations.

You built a great-looking, feature-rich website. But it also has a big problem: it's just too slow. As a result, your visitors are frustrated, and your bounce-rate is high. So now, what do you do? In this session I will explain what web performance actually means, and how to measure the performance of your website. I will show various tools and techniques you can use to pinpoint the performance bottlenecks, and various methods and technologies you can use to overcome them. Finally, I will describe processes you can put in place to ensure that the performance of your website doesn't degrade again in the future.

Something changes in companies when they grow above 150 employees. But it is not only possible to achieve exponential business Growth with a steady head count, it’s also more fun, engaging and effective. Instead of hiring, focus your efforts in Growth dev, marketing and the happiness of your employees.

Reactive Extension (Rx) libraries are everywhere, with the promise that they will make our code more declarative, less prone to bugs and overall simpler. But there is a price to pay for this simplicity – the Rx learning curve is notoriously steep. So before we jump head-first into yet another paradigm that promises to solve all our problems, let’s figure out what exactly there is to gain. In this talk we’ll look at some common front-end problems, and their proposed solutions in RxJS, and try to answer the question - does FRP really make our lives easier?

One of the critical factors for development velocity is software correctness. Our ability to develop and ship new features fast is bounded by our ability to validate several aspects of the change:

• Does the feature meet the requirements?
• How does the feature affect existing code, and how can it affect the production environment?

In this talk, I’ll focus on testing environments: why developers need a self-serve platform to create a full functioning environment on-demand, how such environments should be managed, and how can one restore part of the lost velocity. I’ll cover an internal system we use at AppsFlyer called ‘Namespaces’ that addresses the issue with the help of Mesos.

The motivation for a shared UI components catalog is clear - sharing components and resources, leads to a higher product quality, with significantly lower effort. However, this kind of an effort requires workflow discipline, which usually fails. So, how do you keep UI Components implementation aligned with designs? Automated visual tests are critical to maintain a UI components catalog, the last standpoint which makes sure that nothing changes unintentionally. In this lecture I'll share insights and experience regarding visual tests automation and integrating with a UI/UX designer team, which ensures high quality with minimum effort.

TBD

It's going to be a surprise this year, it's going to be.... interesting... ;)

Rarely Normal or Normally Rare? Anomalies are by definition different and unconventional. A data point that doesn't follow the general rule. The difficulty with defining the nature of the unconventional resonates with all the steps of trying to find it. In this lecture I'll review the art and science of anomaly detection. We will explore classic and deep learning approaches to define and find anomalies with any domain as well as some motivations to look for them in the first place.

While scheduling a limited number of data workflows is a generally manageable task, scaling to hundreds of data workflows with dependencies and diverse job types, requires a substantial customized engineering, complexity, and overall expensive resources. Serverless-based architectures offer an alternative to traditional resource management.

Tomer Levi explains how the data engineering team at Fundbox uses AWS StepFunctions, Docker containers, and Spark to build a live serverless data orchestration platform. Tomer will further describe AWS StepFunctions state machines, their limitations, and how to overcome them by building a custom job scheduling and dependency features. Finally, the talk will illustrate how resource bottlenecks were overcome using Docker containers and AWS Fargate.

We've all suffered from imposter syndrome from time to time. But it turns out imposter syndrome has some really clear patterns, and there are actually a few simple tips and tricks to start appreciating ourselves more. This talk will provide some tools to help you keep calm and focus on your small successes that eventually translate to big successes - similar to Kaizen. And that all this starts with allowing ourselves to be human first and foremost.

Have you ever tried to design a complicated bit of software and ended up with a great big tangled mess? Designing software is hard and formal processes can be frustrating. Is there a better way to approach this task? Mind maps are a fun and intuitive way to brainstorm and organize your ideas on paper or a whiteboard. In this talk I will show how to use mind maps to design software well, convert your thought process into an excellent technical spec, and break it down into workable tasks while leaving lots of room for agility and flexibility.

It was a quiet Sunday at the office. We deployed an innocent looking change to some legacy code. We were happy... for 24 hours. Then our support team got in touch to tell us about clients reporting an issue. It was quickly linked to our, now notorious, patch.

What happened during the following hours is a tale of fellowship, accountability and positive intentions; but also one of SLA definitions we'd forgotten, oncall procedures we ignored and a revert plan we didn't have.

During the session, I'll give a quick explanation of our outage procedure, talk through the events and actions of the “long night” and give you some valuable company-agnostic takeaways.

Viz.ai uses AI to analyze CT scans of stroke patients to facilitate faster detection and treatment. This is a performance critical mission where any failed processing may result in harm to patients (1 minute delay equals the loss of about a week of healthy life to the patient). We leverage AWS GPUs to deploy our deep learning algorithms, and use dynamic allocation to control costs as the traffic changes throughout the day. Ensuring the health of a AWS GPU instance is critical.

In this postmortem we will tell the story of how we detected a faulty GPU hardware on an AWS instance, and how we now automatically detect and mitigate such GPU hardware failures to ensure 100% uptime.

In our day to day jobs most of us work with people from the other side of the Atlantic, and yet we still haven’t found the perfect way to communicate with our American counterparts. When they say something they often mean something else, and when we Israelis speak, we often come across as insensitive. So how how can you flourish in conversations when you’re from a different culture, and worse yet, speaking in English?

I was born and raised in a very Israeli household in Pittsburgh. Having worked as an American among Israelis and as an Israeli among Americans, I have seen the miscommunication coming from both sides. This talk will include practical methods on how to understand each other better, choose your words more carefully, and have productive conversations.

What if I told you that not all managers know everything?

This lightning talk shows my feeble attempts as a technical engineer manager to not be overwhelmed by my smart minions. In addition, the session talks about of how a manager can bring value even without taking a pause to get a P.hd in every field the team members are experts in.

Embeddings in general and Word embeddings, in particular, are real-valued entity representations able to capture context semantics and trained on unique corpora. Models proposing these representations have gained popularity in recent years, but the issue of the most adequate evaluation method still remains open. I will present an extensive overview of the field of words and non-NLP embeddings evaluation, highlighting main problems and proposing a typology of approaches to evaluation.

Even if you are familiar with React Native, do you really know how it works? Did you know about the bridge which is the secret sauce of React Native and makes it work on every platform and gives you the option to create a Native-Like apps? In my talk,I am going to reveal that there is no magic in React Native. There is a real smart mechanism that lets the JS code communicate with the Native code. A major part of this mechanism is The Bridge which is written in C++ (wait..what??) and mapping between all you app modules and even lets you create custom ones of your own! In Addition I will talk about its new architecture. If you are a curious React developer who believes that knowing the internals of a library makes you a better developer, join me to the journey Under The Bridge.

Everybody talks about cracking the coding interview. Dozens of articles and heated discussions on how to prepare, what to ask, and even how to handle the infamous brainteasers.

But let’s face it: the way we interview engineers is broken, and there is a very good reason for that - we are engineers! We know how to write code, and fix problems, a whole other set of skills than assessing other peoples' technical competence. In this talk, we will share six mistakes we were doing while recruiting engineers and share the path to master the skill nobody talk about - being a better interviewer.

Time estimations are an essential part of the software development process. The rise of agile methodologies has even made it easier to come up with more accurate time estimation for engineering tasks. This talk will highlight subtleties, hidden trap doors, and even paradoxes, that await teams that try to over do time estimations. These include:

• time-to-market vs. calendar deadline
• engineering estimation vs. end-to-end estimation
• why is striving to make fully accurate time estimations dangerous to your business?
• are there project types for which time estimations are not as important?
• what is the "black market of time"?

What were the main challenges of Beresheet's design, and what surprised us during the voyage to the Moon? What were the rejected versions of the spacecraft and how did we choose the final one? In this talk you will hear for the first time about some of the behind-the-scenes decision making moments that eventually made the mission a huge success

Bringing an idea from a poc to production is one of the main challenges data scientists struggle with today. While there’s yet no consensus solution for this, recently more and more leading companies began building data science infrastructures. These are designed to support the cycle of research, deploy, retrain and monitor. My team at Check Point is currently focusing on building such infrastructure for company-wise use, while keeping in mind various use-cases, different data types and separate production environments. In this talk, I’ll share our work done in this area, and our future vision for this infrastructure, and how we tend to build it.

“Just leave the garbage outside and we will take care of it for you”. This is the panacea promised by garbage collection mechanisms built into most software stacks available today. So, we don’t need to think about it anymore, right? Wrong! When misused, garbage collectors can fail miserably. When this happens they slow down your application and lead to unacceptable pauses. In this talk we will go over different garbage collectors approaches in different software runtimes and what are the conditions which enable them to function well.

Deep Learning models have been gaining increasing attention in the recommendation systems community, replacing some of the traditional methods. The sparse nature of the problems and the different inputs types offer unique challenges for feature engineering and architecture planning, in order to balance between memorization and generalization.

During the past 2 years the algorithms team in Taboola moved all of our algorithms to DL. In this talk we'll share the lessons we learned doing so. We'll talk about building NN with multiple input types (click history, text and pictures); feature engineering in DL; capturing interactions between features; and the way modelling decisions are related to system engineering and research culture.

Hiring? So is everyone else. How are you going to make sure you get the best minds on your team? You train them! Yup, your read that right, Mentorship is making a comeback.

While experienced workers may be easier to onboard fast, in today's startup economy there are clear benefits to hiring less experienced (potentially qualified) people who are just getting started. Mentorship has its challenges, but Junior Techies are more available, cost less and don't bring their wrong old habits to work.

A mentorship program can be painlessly applied with a few simple tricks, to help your understaffed team as well.

Did you ever get frustrated spelling a password over the phone? "*Capital* B, and then the *digit* 6, not s-i-x". After making sure the language is not set to Hebrew, capslock is off and your dad can tell a zero from an O, you give up before getting locked out. Now, imagine this is how internet is.

15% of the world’s population experience some form of disability, many of whom consume the web using assistive technologies that basically function as a middleman. Often, due to lacking implementation of websites, these technologies supply a partial, sometimes unusable, experience.

In this session I will give a glance at how web is perceived with assistive technologies, explain how they technically work, and share some practical tips to make your product accessible.

It's everywhere! companies are offering "data driven" culture instead of young and vibrant atmosphere. Statisticians are data scientists, ETL developers - data engineers, data architects appear everywhere (they were here all the time, just feel safe to come out to the light now), and executives are talking about "data assets". One graduate even described himself as "passionately data driven" in his CV!

Has everyone lost their minds?! (Sorry, "got their feature set mixed up")

In this talk I'll share insights I collected during my journey from "plain" software engineering to managing Kenshoo's Data Platform Group. I'll try to demystify the term "data driven" and what it actually means for a company. And most importantly, what to do when your CEO ask you to leverage the power of your data.

We will be examining an extension I’ve recently built for VSCode called Glean. Glean, is an open-sourced refactor toolkit for React apps, and has gained popularity lately, with over 24K downloads. https://marketplace.visualstudio.com/items?itemName=wix.glean Blog: https://www.wix.engineering/blog/finally-a-react-refactoring-tool-introducing-glean

SOA has been around for decades, and its latest iteration - microservices - for a while now. Just five years ago microservices were hip, dominating the agenda at conferences; now we almost take them for granted. With microservice-focused conference talks losing steam, the time is ripe to consider what we, as an industry, can learn from the microservice disruption; more importantly, we have a golden opportunity to consider what we have yet to learn!

In this talk I'll analyse the lessons hammered into us as we've adopted microservices, and take a hard look at the challenges I believe we still struggle with as an industry. If you're not doing microservices you'll leave the room knowing where the pains will come from; if you are, you will leave with ideas and leads on mitigating these pains.

Building chrome extension is so simple & fun. In this talk I will explain the basics of building your first chrome extension, in just a couple of minutes! We will then speak about the various ways to turn your awesome extension to a malicious one. The main purpose here is not to turn you into a hacker, but instead to understand the risks of extensions, and to increase awareness to these "small" and "harmless" plugins.

In the past 10 years as a front end developer, I found myself creating internal management tools for RESTful APIs over and over again, while in fact, the usage of these tools remains the same: creating, updating, and viewing data based on an API.

I had enough of it, so I created RESTool - a generic UI tool for managing RESTful APIs.

In this talk I'll present how to configure, run, and use RESTool, and what caused it to jump from 0 to 170 stars on Github in couple of months.

Link: https://github.com/dsternlicht/RESTool

"Micro Frontends" is the new buzzword in the Frontend world, but too many times people use it in the wrong context or with different things in mind. Micro Frontends can refer to different kinds of solutions that solve different types of problems - starting from using different UI frameworks on the same app or letting different teams work on separate parts of the code independently. In this session, we'll go over the different problems we have when building a big app and how different micro-frontends solutions can help with this.

When I started my army service in Mamram, I saw for the first time a giant mainframe computer - the IBM 370. It contained 1.6 MB of memory, divided into 8 slots of 200KB, one program runs in each one. That was 40 years ago. Since then the world of programming has changed dramatically, but some lessons can be learnt from the old days: can you imagine writing your program on a piece of paper, then printing it into a bunch of punched cards? Or that following an exception you'd get a hexadecimal memory dump, on paper (!), and this was your only debugging tool? Working in the industry today, I can sometimes hardly believe it myself. Software design was key at that time, dry runs and a lot of thought before actually running anything live. Is any of it still relevant today? Let's find out!

Clarizen is a leading SaaS company in project management and work collaboration systems which serve thousands of enterprises around the world. System stability and performance are on our top priority, that is why when complaints about sporadic slowness issues began to appear in the middle of 2018 we immediately started to investigate and track it to wild garbage collections that were happening exactly every four hours. Like Murphy predicted, the problem reached a critical point exactly in the holiday season (Thank you Murphy :-/). In this session you will learn the tools and techniques Clarizen used to analyze the problem and find the surprising reason for the mysterious problem.

Discussing the tools we use as software developers is a normal part of software developer culture. It is normal to dissect the inner workings of a debugger, a compiler or the framework of choice of the moment for the purpose of learning to get more from it. This session is devoted to doing just that to the one tool we ALL use, regardless of programming language or environment but hardly give it any thought - our own mind.

We will discuss the cognitive process of debugging, turning debugging unto itself and gain a glimpse of what we do mentally, emotionally and cognitively when we debug. In the process we will learn how to be better debuggers, programmers and maybe, better human beings.

The field of Data Science is currently on fire. Many people want to work on something related to it. But what does it mean to be working in this field? I am going to show you a part of my daily work as a Machine Learning researcher. Specifically, how do you approach a research paper and implement the described algorithm? In this talk, I will share practical tips and advice I collected during my years in academia and the industry. I will present examples how they can be applied on deep learning algorithms.