<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[the-stepstone-group-tech-blog - Medium]]></title>
        <description><![CDATA[The Stepstone Group is one of the world’s leading job platforms. We create opportunities at scale for jobseekers and companies by leveraging our data, platform, and technology. We do this across a portfolio of more than 20 brands and businesses across more than 30 countries. - Medium]]></description>
        <link>https://medium.com/stepstone-tech?source=rss----f042839a21a9---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>the-stepstone-group-tech-blog - Medium</title>
            <link>https://medium.com/stepstone-tech?source=rss----f042839a21a9---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 23:57:21 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/stepstone-tech" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Deploying LLMs in AWS for job listing summarisation]]></title>
            <link>https://medium.com/stepstone-tech/deploying-llms-in-aws-for-job-listing-summarisation-8c009a9bed70?source=rss----f042839a21a9---4</link>
            <guid isPermaLink="false">https://medium.com/p/8c009a9bed70</guid>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Tim Elfrink]]></dc:creator>
            <pubDate>Mon, 12 Aug 2024 08:45:34 GMT</pubDate>
            <atom:updated>2024-08-12T08:45:34.346Z</atom:updated>
            <content:encoded><![CDATA[<p>It is not only OpenAI GPT models that glitter, and even more so when it comes to summarisation. That is the journey that we embarked on a few months ago when we started to work on the topic of job listing summarisation.</p><p>StepStone, one of the world’s leading job platforms, has for a long time built products revolving around data and AI is at the forefront of its strategy. One of our largest sources of data is job listing texts, and there are many reasons for which their summarisation is important for StepStone, namely:</p><ul><li>Allows users to preview the most important aspects of a job listing, permitting them to sneak peek whether they could be a good fit.</li><li>Unifies writing style, which along a concise and clear text eases readability for job applicant that want a quick overview on the different existing offers.</li><li>Reduces computational time for calculating embeddings, useful for other use cases such as recommendation algorithms for finding suitable job offers for our users.</li><li>Provides a more digestible format for mobile application users.</li></ul><p>As such and given the different requirements existing for the different use cases along the organisation LLMs are a great solution. However, from the very beginning we have put our focus on open-source models or, alternatively, available through AWS.</p><p><strong>What constitutes a good summary</strong></p><p>There are multiple use cases that could use job listing summaries across StepStone, such as chat bots, mobile app, or SEO pages. Thus, each of these uses hold different lengths and requirements for which multiple standards are existing. We can, however, fix some general guidelines to assess the quality of a summary:</p><ul><li>Whether it gathers information on all relevant job listing aspects, being these at least an introduction to the role, requirements, duties, and compensation.</li><li>The existence of factual errors or hallucinations in them, as it is critical not to have any of these being shown to end users.</li><li>Whether they are well-formed and written, containing no unfinished sentences, and maintaining a cohesive style across the whole summary.</li><li>The most valuable information from the listing is contained in the summary. Furthermore, we have used similarity metrics such as BERTScore and BLEURT to control the extractiveness of our summaries with respect to the original listing texts.</li></ul><p>We have counted with the support of StepStone’s Linguistic Services team, that has also provided us detailed feedback on how a good summary should look like.</p><p>Thanks to their support, we have defined the above guidelines and identified what is a high-quality output from our model.</p><p><strong>Objectives and strategy</strong></p><p>Self-hosting models in our own AWS account is a tempting possibility as this would allow to have full control and accountability about them as well as not depending on external sources.</p><p>Open-source models are a great option for this, with Falcon 40B being the first experimented approach that yielded decent results. However, hosting such large models also has disadvantages, particularly when it comes to pricing costs. GPU instances are expensive, and their usage should be carefully planned, as we want no surprises in our billing accounts.</p><p>Using a g5.12xlarge instance for hosting Falcon, would cost <a href="https://instances.vantage.sh/aws/ec2/g5.12xlarge?region=eu-west-1&amp;os=linux&amp;cost_duration=hourly&amp;reserved_term=Standard.noUpfront">$3.573/hour using a 1 year savings plan</a>. This is not even accounting for possible horizontal autoscaling, which considering the current amount of daily new listings coming to StepStone would represent a considerably higher cost than this. Additionally, setting this up in a development environment would double up these costs, for a total spending in the lines of several dozens of thousands of dollars per year.</p><p>In conclusion, we would like to keep costs lower, and by doing this, potentially reduce latency if we jump into a smaller model. We aim to do all of this while preserving or even improving the quality of our summaries.</p><p><strong>Benchmarking models for summarisation</strong></p><p>As mentioned before, serving such large models like Falcon in our own AWS account also means a large bill.</p><p>Strictly looking at performance though, there are a few more things that could be improved from this model’s summaries:</p><ul><li>A limited context length of only 2048 tokens. Considering that job listings can be long, this forced us to tackle such cases with a map reducing approach thanks to LangChain.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/451/1*_q3O4Ss0FGaBQEIHVt_hhQ.png" /><figcaption>Map-reduce approach to summarise long listing texts in Falcon 40B</figcaption></figure><p>This is not ideal as context can get lost and as of such attention could be more sparse than desirable, instead of focusing on the most important parts of the listing.</p><ul><li>Response time is higher than requirements, taking around 15–30 seconds to generate a summary.</li></ul><p>So, before considering other options like model fine-tuning, we would still have to generate high-quality summaries from a model we could distillate knowledge from. This is provoked by the fact that curating manual training samples is an extensive and lengthy process.</p><p>Therefore, our mandatory first step is to ensure the generation of good summaries.</p><p><strong>Testing summary generation with various models</strong></p><p><a href="https://aws.amazon.com/bedrock/">Bedrock</a> is a new service offered by AWS that grants access to foundation models through a simple API call. It gives access to multiple powerful LLMs such as Titan or models from A21 Labs and Anthropic. From these, we have extensively experimented with model like <a href="https://www.anthropic.com/index/claude-2">Claude 2</a>, one of the most capable models of those offered on Bedrock. It allows for a large context length (up to 100,000 tokens), a robust performance and availability, and speedy output generation. Next to closed source models it also allows to use massive open source models like Mixtral 8X7B Instruct which usually is very expensive to host yourself. This advantage by paying per token vs per hour the instance works can be really cost effective.</p><p>Bedrock will be packaged as a Docker image available through AWS Lambda. This way, we can access every Bedrock model through a simple Lambda invocation from any region, provided we can make a HTTP request through an Application Load Balancer (ALB) or API Gateway.</p><p>For generating summaries on demand, a Docker container is created with the above workflow and, to productionise it, we embed it into the following process:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/451/1*c88dii_VIVar9Z7QW--LUQ.png" /><figcaption>Architecture for generating summaries in bulk or on-demand</figcaption></figure><p>Periodically, we run a job as scheduled by a CloudWatch event which will start running a Lambda that starts a Processing job in SageMaker with the above workflow in it. This process can be also run under demand thanks to our ALB. The output summaries can be ingested by other StepStone services or directly dumped into a S3 bucket.</p><p><strong>Operational costs</strong></p><p>While models like Claude and Mixtral proved to be exceptionally performing for summary generation, it can be expensive to maintain an infrastructure relying on Bedrock for our solution. That, and the privacy concerns for the data we use, prompted us to steer away from this approach, and try a cheaper but self-hosted solution from an open-source model.</p><p><a href="https://arxiv.org/abs/2305.11206">As literature has shown that with just 1,000 samples, we can already safely fine-tune LLMs</a>.</p><p><a href="https://www.philschmid.de/sagemaker-llama2-qlora">Let’s try to fine-tune LLaMa 2 7b with QLoRA</a> which quantizes a pre-trained model to just 4 bits and reduces memory usage allowing to run the fine-tuning on a single GPU instance.</p><p>To generate the input dataset, we set up a process with of creating summaries for a subset of our listings, and then re-evaluating them with the support of our Linguistics team. This step of the process was to ensure that the input for the finetuned model is conforming our internal listing text guidelines and tone.</p><p>This task takes just under 2 hours for such small set of samples, which considering the instance pricing it amounts to a total of:</p><p>1.95h * 1.813 $/h = <strong>$3.53 for fine-tuning the model</strong></p><p>So, we can see that it is a cheap task. Furthermore, such model can run on a g5.2xlarge instance, <a href="https://instances.vantage.sh/aws/ec2/g5.2xlarge?region=eu-west-1&amp;os=linux&amp;cost_duration=hourly&amp;reserved_term=Standard.noUpfront">hosted for only $0.8524/hour</a>. This would represent a staggering <strong>76.14% cheaper price than the original Falcon 40B model.</strong></p><p><strong>Conclusion and final takes</strong></p><p>We have compared the performance from different LLMs for different tasks and presented a proposed architecture for generating summaries in bulk in your own AWS infrastructure, while ensuring efficiency in resources and a consistently high quality on all outputs.</p><p>Thank you for reading!</p><p>Special thanks to <a href="https://medium.com/u/5a96788cc0f2"><strong>Ricardo Fernandez</strong></a><strong> </strong>for writing the original post and integrate the solution!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8c009a9bed70" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stepstone-tech/deploying-llms-in-aws-for-job-listing-summarisation-8c009a9bed70">Deploying LLMs in AWS for job listing summarisation</a> was originally published in <a href="https://medium.com/stepstone-tech">the-stepstone-group-tech-blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Automating DynamoDB Backups to S3 using AWS EventBridge Scheduler and Terraform]]></title>
            <link>https://medium.com/stepstone-tech/automating-dynamodb-backups-to-s3-using-aws-eventbridge-scheduler-and-terraform-2edb8fff1914?source=rss----f042839a21a9---4</link>
            <guid isPermaLink="false">https://medium.com/p/2edb8fff1914</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[eventbridge]]></category>
            <category><![CDATA[aws-lambda]]></category>
            <category><![CDATA[dynamodb]]></category>
            <dc:creator><![CDATA[Tim Elfrink]]></dc:creator>
            <pubDate>Wed, 22 Nov 2023 09:44:35 GMT</pubDate>
            <atom:updated>2023-11-22T09:44:35.181Z</atom:updated>
            <content:encoded><![CDATA[<p>Regular backups of your DynamoDB tables are essential for ensuring data reliability, disaster recovery, and compliance. In this blog post, we’ll explore an efficient and automated way to backup your DynamoDB data to Amazon S3 using AWS EventBridge Scheduler and Terraform.</p><h4>Manual Backup with AWS CLI</h4><p>According to AWS <a href="https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/S3DataExport_Requesting.html">documentation</a> you can easily create a one time full backup to S3 with the following CLI command (Thanks Jaroslaw Brodowski for pointing out this feature!) :</p><pre>aws dynamodb export-table-to-point-in-time \<br>  --table-arn arn:aws:dynamodb:us-west-2:123456789012:table/MusicCollection \<br>  --s3-bucket ddb-export-musiccollection-9012345678 \<br>  --s3-prefix 2020-Nov \<br>  --export-format DYNAMODB_JSON \<br>  --export-time 1604632434 \<br>  --s3-bucket-owner 9012345678 \<br>  --s3-sse-algorithm AES256</pre><p>This command is ideal for creating a one-time backup quickly. Despite its utility, remember that manual processes are error-prone and not scalable for regular intervals.</p><h4>Automating with AWS Lambda and EventBridge</h4><p>One common pattern to automate this process is by using an AWS Lambda function to execute the above command and trigger it using AWS EventBridge, as described in <a href="https://dev.to/ritaly/how-to-export-aws-dynamodb-data-to-s3-for-recurring-tasks-4l47">this blog</a>.</p><p>While AWS Lambda functions triggered by EventBridge are a conventional choice for automating backups, they involve additional complexities such as code deployment and monitoring.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/466/0*izVMBogOA4WDXYow.png" /><figcaption>Credits: <a href="https://dev.to/ritaly/how-to-export-aws-dynamodb-data-to-s3-for-recurring-tasks-4l47">https://dev.to/ritaly/how-to-export-aws-dynamodb-data-to-s3-for-recurring-tasks-4l47</a></figcaption></figure><h4>Simplifying with AWS EventBridge Scheduler</h4><p><a href="https://aws.amazon.com/blogs/compute/introducing-amazon-eventbridge-scheduler/">Introduced in late 2022</a>, AWS EventBridge Scheduler offers a more straightforward solution. This tool allows you to call over 270 services and over 6,000 API actions with AWS SDK targets directly. With EventBridge Scheduler, you can eliminate the need for an AWS Lambda function and call the DynamoDB API from a scheduled event.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/326/1*B6nKZbWWn_DtF4rI2NT55w.png" /><figcaption>Schedule directly from EventBridge</figcaption></figure><p>In Terraform, implementing this solution is as simple as:</p><pre>resource &quot;aws_scheduler_schedule&quot; &quot;dynamodb_table_backup_music_collection&quot; {<br>  name = &quot;backup-music-collection&quot;<br><br>  schedule_expression = &quot;rate(30 days)&quot; # or something like &quot;cron(0 0 1 * * *)&quot;<br><br>  targets {<br>    role_arn = &quot;arn:aws:iam::123456789012:role/MusicCollectionBackup&quot;<br>    arn      = &quot;arn:aws:scheduler:::aws-sdk:dynamodb:exportTableToPointInTime&quot;<br>    input = &lt;&lt;EOF<br>            {<br>              &quot;TableArn&quot;: &quot;arn:aws:dynamodb:us-west-2:123456789012:table/MusicCollection&quot;,<br>              &quot;S3Bucket&quot;: &quot;ddb-export-musiccollection-9012345678&quot;,<br>              &quot;ExportFormat&quot;: &quot;DYNAMODB_JSON&quot;,<br>              &quot;S3SseAlgorithm&quot;: &quot;AES256&quot;<br>            }<br>            EOF<br>  }<br>}</pre><p>Ensure that your role_arn has the necessary permissions to assume scheduler.amazonaws.com, export the DynamoDB table, and write to your S3 bucket. Also, note that the DynamoDB table must have point-in-time recovery enabled for this method to work.</p><p>With just a few lines of Terraform, you can set up regular full exports of your DynamoDB table to be stored in your S3 bucket, making your data readily accessible for various use cases.</p><h4>Cost Considerations for DynamoDB Backups</h4><p>Ensuring regular DynamoDB backups is crucial, but it’s important to factor in costs. While automated backups to S3 through AWS EventBridge Scheduler provide long-term storage and comply with many regulatory norms, they may incur higher costs for large or frequent backups. An alternative or complementary cost-effective option is DynamoDB’s Point-In-Time Recovery (PITR), which offers continuous, incremental backups for up to 35 days and can provide significant savings if your restoration needs are within this window. By understanding your data recovery requirements and leveraging the right mix of PITR and S3 backups, you can formulate a backup strategy that balances cost-efficiency with operational resilience.</p><h4>Conclusion</h4><p>Automating DynamoDB backups to S3 is now more accessible than ever, thanks to AWS EventBridge Scheduler. This solution simplifies the process and provides a cost-effective way to ensure data reliability and recovery. By following the steps outlined in this post, you can streamline your backup strategy and focus on leveraging your DynamoDB data from S3. Give it a try and share your experiences with us.</p><p>Read more about <a href="https://medium.com/stepstone-tech/about-thestepstonegroup/home">The Stepstone Group</a>, the<a href="https://medium.com/stepstone-tech/technology/home"> technologies </a>we use or take an inside look at our <a href="https://medium.com/stepstone-tech/organisation-processes/home">organisation &amp; processes</a>. Interested in joining us? Check out our <a href="https://medium.com/stepstone-tech/careers-stepstone-d287c9d0d26d">careers page</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2edb8fff1914" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stepstone-tech/automating-dynamodb-backups-to-s3-using-aws-eventbridge-scheduler-and-terraform-2edb8fff1914">Automating DynamoDB Backups to S3 using AWS EventBridge Scheduler and Terraform</a> was originally published in <a href="https://medium.com/stepstone-tech">the-stepstone-group-tech-blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Representing The Stepstone Group at the Düsseldorf Data Science Meetup]]></title>
            <link>https://medium.com/stepstone-tech/representing-the-stepstone-group-at-the-d%C3%BCsseldorf-data-science-meetup-6dab05ef551e?source=rss----f042839a21a9---4</link>
            <guid isPermaLink="false">https://medium.com/p/6dab05ef551e</guid>
            <category><![CDATA[salary]]></category>
            <category><![CDATA[stepstone]]></category>
            <category><![CDATA[meetup]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Michael Matuschek]]></dc:creator>
            <pubDate>Fri, 29 Sep 2023 11:51:18 GMT</pubDate>
            <atom:updated>2023-10-02T06:56:54.149Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Brought to you by the Data Science and Enrichment Team @ The Stepstone Group</em></p><p>On August 29th, <a href="https://www.linkedin.com/in/timelfrink">Tim Elfrink</a> and myself participated in the Düsseldorf Data Science Meetup to represent The Stepstone Group. We were invited to speak at the trivago headquarters in Düsseldorf, and shared the stage with <a href="https://www.linkedin.com/in/alexander-fischer-25653219b">Alexander Fischer</a> from trivago who presented his work about “Linear Regression (with high dimensional fixed effects) with PyFixest.”</p><p>The topic of our talk was “The World of Salary: Introducing Salary Prediction Models at StepStone”, and we gave an overview of the relevance of salary information in the job market, the salary-related products at Stepstone and the technical solutions that make them happen.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/575/1*u2RdL56jvlZGIvFo5FBEzg.png" /></figure><p>Specifically, I talked about how we collect salary data through the Salary Planner, how we leverage the data for products like Salary on Listings, and what impact this has on our business. I also dived deeper into the Data Science pipeline, including the data cleaning and filtering, feature engineering and model tuning. If you want to learn more about this, I can also recommend <a href="https://medium.com/stepstone-tech/the-world-of-salary-introducing-salary-prediction-models-at-stepstone-5d7f13e33fa6">this piece</a> by our teammate <a href="https://www.linkedin.com/in/klea-sanka">Klea Sanka</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/394/1*xYycgN2194gwapicZKGhBA.png" /></figure><p>Tim then took over to give an overview of our Machine Learning infrastructure, and explained how we use AWS SageMaker to make the whole process of collecting the data, training the models and deploying them to production highly automated, scalable and robust, accommodating different technical and business requirements.</p><p>The talk was extremely well received, and we got a lot of thoughtful and interesting questions regarding the dynamics of salary data, our models and infrastructure, but also more general questions about salary (for instance from HR people) as this is of course a very hot topic for everyone in the job market.</p><p>Even long after the talk, over pizzas and beer, we were still tied up in discussions about the topic, and people reached out to us afterwards with more questions still.</p><p>It was a lovely evening all in all, and not only do I need to thank my colleague Tim and co-presenter Alex, but also <a href="https://www.linkedin.com/in/aidaorujova">Aida Orujova</a> from trivago who organized everything meticulously and made us feel right at home. We will be happy to meet again at one of the next Data Science meetups in Düsseldorf!</p><p><a href="https://medium.com/stepstone-tech/deploying-and-fine-tuning-llms-in-aws-for-job-listing-summarisation-c3e16e00aab3">Here</a> you can learn about another cool use case from our team using Large Language Models.</p><p>You can also read more about <a href="https://medium.com/stepstone-tech/about-thestepstonegroup/home">The Stepstone Group</a>, the<a href="https://medium.com/stepstone-tech/technology/home"> technologies </a>we use or take an inside look at our <a href="https://medium.com/stepstone-tech/organisation-processes/home">organisation &amp; processes</a>.</p><p>Interested in joining us? Check out our <a href="https://medium.com/stepstone-tech/careers-stepstone-d287c9d0d26d">careers page</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6dab05ef551e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stepstone-tech/representing-the-stepstone-group-at-the-d%C3%BCsseldorf-data-science-meetup-6dab05ef551e">Representing The Stepstone Group at the Düsseldorf Data Science Meetup</a> was originally published in <a href="https://medium.com/stepstone-tech">the-stepstone-group-tech-blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Stepstone Group at RANLP 2023]]></title>
            <link>https://medium.com/stepstone-tech/the-stepstone-group-at-ranlp-2023-c1b5d3296c26?source=rss----f042839a21a9---4</link>
            <guid isPermaLink="false">https://medium.com/p/c1b5d3296c26</guid>
            <category><![CDATA[conference]]></category>
            <category><![CDATA[bias]]></category>
            <category><![CDATA[stepstone]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[Vera Aleksic]]></dc:creator>
            <pubDate>Fri, 22 Sep 2023 11:16:03 GMT</pubDate>
            <atom:updated>2023-09-22T11:16:03.012Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/846/1*JO9jSvmp5OG2D9nDZglO1Q.jpeg" /></figure><p>In the middle of my summer vacation, I took a detour to attend and present at the 14th RANLP Conference (Recent Advances in Natural Language Processing, <a href="https://ranlp.org/ranlp2023/">https://ranlp.org/ranlp2023/</a> ) and to visit the beautiful city of Varna on the Black Sea coast in Bulgaria.</p><p>With a history spanning over more than two decades, this important conference brings together NLP researchers and professionals from all over the world and offers them a forum for discussing the latest developments and innovations in NLP. What I liked very much was learning a lot about the advances in AI and ML, but also seeing the diversity and depth of contributions in some more traditional research fields such as phonetics, morphology, syntax, discourse, pragmatics, lexicography.</p><p>Our paper was just one of 140 accepted research papers this year, which will be published in the remarkable 1300-pages proceedings file (<a href="https://ranlp.org/ranlp2023/index.php/proceedings/">https://ranlp.org/ranlp2023/index.php/proceedings/</a>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*G3OY_MnI5g4f14BOoRT3lQ.jpeg" /></figure><p>The conference featured six distinguished keynote speakers, each renowned in their respective NLP subfields (<a href="https://ranlp.org/ranlp2023/index.php/keynote-speakers/">https://ranlp.org/ranlp2023/index.php/keynote-speakers/</a>). I love listening to keynote speeches and I never miss them at conferences! Good keynote speakers not only share insights into various aspects of their research, but also provide a human and empathetic perspective. They enrich their presentations with philosophical insights and remind us that the true essence of NLP lies not only in algorithms and data but also in its impact on the way we communicate and understand each other. So did the RANLP keynote speakers!</p><p>As expected, one of the most important topics was the advancement of AI, in particular the emergence of large generative models (ChatGPT, Bard, PaLM…). On the one hand, great admiration for the quality of LLMs was expressed (“Does God speak to them?”) and extensive research on how to use, adapt and improve them was presented. As the main directions in the “New NLP” researchers see making LLMs <strong>usable</strong> (prompt engineering, smaller and cheaper LLMs), <strong>scientific</strong> (understand how they do what they do), and <strong>useful</strong> (tune them to domains, tailor them for enterprise processing). On the other hand, many large-scale tests and analyses aim at finding errors and limits of LLMs (Do they “know” and “understand” numbers, facts, text?). In this context, the conference continued the global discussion on the potential risks and uncertainties associated with their use. Also, ethical questions were raised, with a strong emphasis on our ethical responsibility as researchers and professionals when dealing with the ‘all-powerful’ machine.</p><p>Unsurprisingly, not only the primary focus of some keynotes, but also of many of the submitted papers, revolved around the topics like bias and hate speech — how to detect them and how to effectively address and remove them from the digital landscape. Most researchers struggle with data scarcity, especially how to cover smaller languages, all sub-cultures, all biases… But nowadays it seems more difficult than ever to first define what is bias, what is hate speech, and then how to fairly annotate them (“Who decides? How many people does it require to call a text hate speech? The majority? A single person?”).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oCOvErdIbrgtM8rYD0ejtw.jpeg" /></figure><p>The main conference programme was organised in 12 parallel presentations sessions and two poster sessions, with topics such as Language Resources, Language Models, Machine Translation, Sentiment Analysis, Summarisation, Generation, and many more.</p><p>The Stepstone Group was represented with a contribution from the Linguistic Services Team. Our submitted paper outlined an automatic sentence generation pipeline, developed last year for the Stepstone Recruit project. Internally, we refer to it as “linguistic generation” serving as the backup to the AI-based sentence generation model created by the data scientists of the Novel Team.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YH3B2fYBSp54_JI9p_5EdA.jpeg" /></figure><p>Feel free to check out the paper as soon as the proceedings are published:</p><p>Vera Aleksic, Mona Brems, Anna Mathes, Theresa Bertele. 2023. Lexicon-driven automatic sentence generation for the skills section in a job posting. <em>Proceedings of Recent Advances in Natural Language Processing</em>, pages 32–40. Varna, Sep 4–6, 2023. <a href="https://doi.org/10.26615/978-954-452-092-2_004">https://doi.org/10.26615/978-954-452-092-2_004</a></p><p>Lastly, I believe that the organisers found it easy to plan a captivating social program and exciting excursions, by having the breathtaking backdrop of the Black Sea, the picturesque city of Varna, as well as many archaeological sites at their disposal. A place to visit again! 😊</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RlAmYHvB9CB71sREcxzgKA.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c1b5d3296c26" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stepstone-tech/the-stepstone-group-at-ranlp-2023-c1b5d3296c26">The Stepstone Group at RANLP 2023</a> was originally published in <a href="https://medium.com/stepstone-tech">the-stepstone-group-tech-blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Register external services containing paths in endpoints into Consul Service Mesh]]></title>
            <link>https://medium.com/stepstone-tech/register-external-services-containing-paths-in-endpoints-into-consul-service-mesh-d6c7b67e4eb3?source=rss----f042839a21a9---4</link>
            <guid isPermaLink="false">https://medium.com/p/d6c7b67e4eb3</guid>
            <category><![CDATA[service-mesh]]></category>
            <category><![CDATA[stepstone]]></category>
            <category><![CDATA[consul]]></category>
            <category><![CDATA[terminating-gateway]]></category>
            <category><![CDATA[nomad]]></category>
            <dc:creator><![CDATA[Bartłomiej Więckowski]]></dc:creator>
            <pubDate>Fri, 22 Sep 2023 08:21:42 GMT</pubDate>
            <atom:updated>2023-09-27T11:23:32.873Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/828/1*gsKpaqYz5QqnwT96UJY2MQ.jpeg" /></figure><h3>Problem</h3><p>When you deploy an application into a Consul Service Mesh it will deploy together with sidecar envoy proxy that handles your network traffic. This application can only access services inside mesh. One way to access any service outside the mesh is to register it using a <a href="https://developer.hashicorp.com/consul/docs/connect/gateways/terminating-gateway">Terminating Gateway</a>.</p><p>It some cases the external application you want to register into Service Mesh can be hosted behind another proxy under specific path i.e your-proxy.domain.com/path/to/service. It may seem to be pretty easy at first glance but putting the whole address into the <a href="https://developer.hashicorp.com/nomad/docs/job-specification/service#address">Consul Service address</a> field will not work and cause a No health upstream error message.</p><p>Sometimes there is a general misunderstanding on how envoy proxy servers route traffic. If you are attempting to route the incoming path / to a backed service path of /some/path, this is not how it works. Envoy is a L3/L4 network proxy so it is not able to interpret paths and your incoming endpoint will be passed along to the backend service without any paths.</p><h3>Prerequisites</h3><p>In order to test this solution you need to have:</p><ul><li>Consul Service Mesh</li><li>Nomad Cluster</li><li>Network connectivity to external application</li><li>Terminating Gateway deployed in Service Mesh</li></ul><h3>Solution</h3><p>One solution is to use the Consul KV store and the Nomad job templating mechanism.</p><p>We will create a new environment variable for mesh service to use it as an endpoint to an external service.</p><p>There are multiple ways of interacting with Consul API. In this article we will use Terraform.</p><h4>Registering service</h4><p>It is vital to create a Consul KV store entry along with a Consul Service as we will use it in next steps.</p><pre>resource “consul_service” “this” {<br>  name = var.external_service.name<br>  node = &quot;&lt;consul_node_name&gt;&quot;<br>  port = var.external_service.port<br>  <br>  check {<br>    check_id        = var.external_service.name<br>    name            = “${var.external_service.name} health check”<br>    status          = “passing”<br>    http            = “https://${var.external_service.addresses}${var.external_service.path}${var.external_service.health_check_path}&quot;<br>    tls_skip_verify = false<br>    interval        = “30s”<br>    timeout         = “3s”<br>  }<br>}<br><br>resource &quot;consul_key&quot; &quot;this&quot; {<br>  datacenter = &quot;&lt;your-consul-data-center&gt;&quot;<br>  <br>  key {<br>    path  = &quot;external-services/${var.external_service.name}/path_to_service<br>    value = var.external_service.path<br>  }<br>}</pre><h4>Consuming endpoints in the mesh</h4><p>Once we have our service registered and key in the Consul KV store created we can make use of it in our Nomad job.</p><p>To make a connection possible we need to first register the external service as an upstream in our mesh. We can pass a list of upstreams to our Nomad job definition.</p><pre>variable &quot;upstreams&quot; {<br>  type    = list(string)<br>  default = [&quot;external-service-a&quot;, &quot;external-service-b&quot;]<br>}<br><br>resource &quot;nomad_job&quot; &quot;this&quot; {<br>  ...<br>  hcl2 {<br>    ...<br>    upstreams = var.upstreams<br>    ...<br>  }<br>}</pre><p>In our job.nomad.hcl we will use the Nomad job templating mechanism and Nomad built-in environment variables to create a new external service endpoint.</p><pre>variable &quot;upstreams&quot; {<br>  type = list(string)<br>}<br><br>locals {<br>  upstream_starting_port = 12345<br>  upstreams              = join(&quot; &quot;, [for upstream in upstreams : &quot;\&quot;${upstream}\&quot;&quot;])<br>}<br><br>job &quot;name&quot; {<br>  ...<br>  group &quot;name&quot; {<br>    service &quot;name&quot; {<br>      dynamic &quot;upstreams&quot; {<br>        for_each = [for index, upstream in var.upstreams : {<br>          destination_name = upstream<br>          local_bind_port  = (local.upstream_starting_port + index)<br>        }]<br>  <br>        content {<br>          destination_name = upstreams.value.destination_name<br>          local_bind_port  = upstreams.value.local_bind_port<br>        }<br>      }<br>    }<br>    ...<br>    task &quot;name&quot; {<br>      ...<br>      template {<br>        data = &lt;&lt;EOF<br>{{- range $u := sprig_list ${local.upstreams} }}<br>EXTERNAL_APPLICATION_URL_{{ $u | replaceAll &quot;-&quot; &quot;_&quot; }}=http://{{ $u | sprig_cat &quot;NOMAD_UPSTREAM_ADDR_&quot; | sprig_nospace | env }}{{ keyOrDefault (sprig_cat &quot;external-services/&quot; $u &quot;/path_to_service&quot; | sprig_nospace) &quot;&quot; }}<br>{{- end -}}<br>        EOF<br>        destination = &quot;local/env_vars&quot;<br>        env         = true<br>      }<br>    }<br>  }<br>}   </pre><p>We had to create an additional local.upstreams variable to fit the format of the list that is required by Go templating.</p><p>Next we iterated over our list of upstreams and we templated new environment variables that use <a href="https://developer.hashicorp.com/nomad/docs/runtime/environment">Nomad Runtime environment variable</a> <em>NOMAD_UPSTREAM_ADDR_&lt;service_name&gt;</em>.</p><p>We also used a several <a href="https://github.com/hashicorp/consul-template/blob/main/docs/templating-language.md">Consul Templating functions</a> to manipulate data such as:</p><ul><li>range</li><li>env</li><li>replaceAll</li><li>sprig_list</li><li>sprig_cat</li><li>sprig_nospace</li><li>keyOrDefault</li></ul><p>Have you come across this before? What solution have you used? If not, I hope it was helpful for you.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d6c7b67e4eb3" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stepstone-tech/register-external-services-containing-paths-in-endpoints-into-consul-service-mesh-d6c7b67e4eb3">Register external services containing paths in endpoints into Consul Service Mesh</a> was originally published in <a href="https://medium.com/stepstone-tech">the-stepstone-group-tech-blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The potential of augmented reality]]></title>
            <link>https://medium.com/stepstone-tech/the-potential-of-augmented-reality-508bd7edca7?source=rss----f042839a21a9---4</link>
            <guid isPermaLink="false">https://medium.com/p/508bd7edca7</guid>
            <category><![CDATA[virtual-reality]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[stepstone]]></category>
            <category><![CDATA[augmented-reality]]></category>
            <dc:creator><![CDATA[Mazen Mezher]]></dc:creator>
            <pubDate>Wed, 23 Aug 2023 09:56:20 GMT</pubDate>
            <atom:updated>2023-08-23T09:56:20.229Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/904/1*h1x_HhxpysQW-6ROApdhfA.png" /></figure><p><em>The text you are about to read has been modified using the Open AI GPT.</em></p><p>The captivating realm of augmented reality (AR) and virtual reality (VR) has been a driving force behind my curiosity for years. As an enthusiast who has personally navigated their evolution, delving into both usage and development, I am genuinely fascinated by the rapid strides these technologies have taken since the beginning. Together, we embark on an exploration of the limitless possibilities they bestow upon our ever-expansive digital landscape, driven by an innate motivation to unveil the captivating future they hold.</p><p>Augmented reality (AR) is a revolutionary technology that’s been around for several decades. But did you know that its origins can be traced back to the 1960s? That’s right! A computer scientist named Ivan Sutherland was one of the pioneers of computer graphics and had a vision to create a “virtual reality” system that would allow users to interact with a 3D computer-generated environment in real-time.</p><p>Augmented reality (AR) and virtual reality (VR) are two different technologies today that change the way we interact with digital content. While both have the goal of enhancing our experience with technology, they do it in very different ways. AR enhances our real-world environment by overlaying digital content on top of it, creating a mixed reality experience. It adds digital elements to our actual surroundings, which allows us to interact with them in a more immersive way.</p><p>On the other hand, VR completely immerses us in a digital world, blocking out our physical environment entirely. It creates a simulated environment that feels like a real one, allowing us to experience things that would be impossible in the real world. Both technologies have unique advantages and disadvantages. AR can be more practical and accessible since it does not require any additional hardware besides a smartphone or tablet. It can also be used in a variety of industries, from gaming and entertainment to healthcare and education.</p><p>Meanwhile, VR provides a more complete and immersive experience, making it ideal for industries like gaming and architecture, where a detailed simulation is necessary. However, it requires specialized hardware like headsets and controllers, which can be expensive and not as widely accessible. While AR and VR have different approaches to enhancing our digital experiences, neither is better or worse both have their own unique benefits and limitations.</p><p>However, it was not until the early 1990s that the term “augmented reality” was coined for the technology that overlays digital information onto of the real world. A Boeing researcher named Tom Caudell was exploring ways to improve productivity on the factory floor and realized that AR technology could be used to overlay digital information onto physical objects to help workers perform tasks more efficiently.</p><p>Fast forward a few decades, and AR has become more accessible to everyday users thanks to advancements in hardware and software. But it was not until the introduction of the iPhone in 2007 that AR really began to take off. With smartphones rapidly increasing in popularity, AR became more accessible to the masses.</p><p>Since then, AR on smartphones has come a long way. Today, there are millions of users around the world who regularly use AR apps on their smartphones. From gaming and entertainment to education and healthcare, AR is now being used for a wide variety of purposes. One app in specific that really put AR on the map was Pokémon Go. This app has passed the 1 billion download mark.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/904/1*fGdLRdFVOGiBeDYYx0wBlQ.jpeg" /></figure><p>With AR, you can use your smartphone or other devices that has a camera and sensors combined with AR software to be able to see digital objects and information superimposed onto the world around you. It is like having a virtual layer on top of reality! AR will have an important role in the future when companies make really good AR glasses. I think it will become an extra tool for people to use just as their phones to easily get information on the go while they are out and about.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S9WDKXrQia2pfLPFoTfXBw.jpeg" /></figure><p>However, one of the key advantages of AR on smartphones right now is that it allows users to experience the technology without the need for specialized hardware like AR glasses that are a bit further away even though they exist today they are not popular enough to be used as commonly as smartphones. Today all you need is a smartphone and an AR app, and you can begin exploring the endless possibilities of AR right away. And with smartphone manufacturers like Apple and Google investing heavily in AR capabilities, the technology is only going to get better and more immersive.</p><p>After considering all the information presented, it appears that AR technology has great potential for the future. It is uncertain whether it will become widely adopted, but regardless, there are many current applications for AR on smartphones that can benefit a diverse range of users. For example, gamers can enhance their gameplay experience, students can find new and creative ways to learn, and technology enthusiasts can explore new possibilities. If AR continues to gain popularity, there may even be new and better AR devices introduced to the market. So why not take the opportunity to explore this fascinating world of augmented reality by downloading an AR app on your smartphone today?</p><p>Read more about <a href="https://medium.com/stepstone-tech/about-thestepstonegroup/home">The Stepstone Group</a>, the<a href="https://medium.com/stepstone-tech/technology/home"> technologies </a>we use or take an inside look at our <a href="https://medium.com/stepstone-tech/organisation-processes/home">organisation &amp; processes</a>. Interested in joining us? Check out our <a href="https://medium.com/stepstone-tech/careers-stepstone-d287c9d0d26d">careers page</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=508bd7edca7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stepstone-tech/the-potential-of-augmented-reality-508bd7edca7">The potential of augmented reality</a> was originally published in <a href="https://medium.com/stepstone-tech">the-stepstone-group-tech-blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Beginner’s Guide to Crafting Effective Prompts for ChatGPT]]></title>
            <link>https://medium.com/stepstone-tech/a-beginners-guide-to-crafting-effective-prompts-for-chatgpt-3c54672a5fc4?source=rss----f042839a21a9---4</link>
            <guid isPermaLink="false">https://medium.com/p/3c54672a5fc4</guid>
            <category><![CDATA[guides-and-tutorials]]></category>
            <category><![CDATA[openai-chatgpt]]></category>
            <category><![CDATA[prompt]]></category>
            <category><![CDATA[stepstone]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[Thiemo Bubel]]></dc:creator>
            <pubDate>Fri, 18 Aug 2023 14:47:54 GMT</pubDate>
            <atom:updated>2023-08-21T05:54:02.046Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XPgYTtRQ6ChCv43i" /><figcaption>Photo by <a href="https://unsplash.com/@iammottakin?utm_source=medium&amp;utm_medium=referral">Mojahid Mottakin</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Hello there, tech enthusiasts and curious minds! Have you ever wondered how to make the most out of ChatGPT, that nifty AI language model that can chat like a real human? Well, you’re in luck! Today, we’re diving into the art of creating the perfect prompt to get the responses you’re looking for. Don’t worry, we’ll break it down step by step, in the most friendly and pproachable way possible. Let’s get started!</p><h3>**1. Set the Scene with Context:</h3><p>Imagine you’re chatting with a new friend — you wouldn’t just start with “Hey, tell me something.” You’d probably say something like “Hey, tell me about your favorite hobbies.” In the same way, give ChatGPT a bit of context to kick off the conversation right.</p><p>Example:</p><pre>You&#39;re an AI travel buddy, helping me plan a dream vacation to Japan.<br>What are some must-visit places and local dishes I should try?</pre><h3>**2. Be the Director: Give Clear Instructions:</h3><p>Think of ChatGPT as your helpful assistant who’s eager to help but needs a bit of direction. Be specific about what you’re asking for, so it knows exactly how to assist you.</p><p>Example:</p><pre>Pretend you&#39;re a fitness guru. Design a simple workout routine for beginners, <br>including warm-up exercises and cool-down stretches.</pre><h3>**3. Picture the End Result: Desired Format:</h3><p>Decide how you want your answer. Do you need a list, a short paragraph, or maybe a back-and-forth conversation? This helps ChatGPT organize its thoughts to match what you’re looking for.</p><p>Example:</p><pre>Give me three quick and healthy breakfast recipes along with the step-by-step<br>instructions for each.</pre><h3>**4. Keep the Conversation Flowing with Follow-ups:</h3><p>We all love it when conversations flow naturally. If you want to build a chat-like interaction, use follow-up prompts that refer back to what ChatGPT just said. It’s like having a chat buddy who remembers what you talked about!</p><p>Example:</p><pre>User: Which movies are topping the charts right now?<br>AI: Currently, movies like &quot;StarQuest&quot; and &quot;Adventure Awaits&quot; are making waves. <br>Want to know more about their plots?</pre><h3>**5. Tweaking for Perfection: Iterative Refinement:</h3><p>No need to stress if the first response isn’t quite what you had in mind. Give your prompt a little makeover! Adjust the wording, add details, or refine your question until you get the answer you’re looking for.</p><p>Example:</p><pre>Prompt 1: Suggest some sci-fi books.<br>Prompt 2: Recommend classic sci-fi novels with mind-bending time travel plots<br>that&#39;ll leave me questioning reality!</pre><h3>**6. Add a Pinch of Personality: Politeness and Persona:</h3><p>You can even give ChatGPT a touch of personality! Just like how you’d talk differently to a teacher and a best friend, you can set the tone and style of your AI’s responses.</p><pre>Imagine you&#39;re a cheerful science geek. Explain the concept of black holes <br>in a way that even I can understand!</pre><h3>**7. Length Matters: Length and Depth:</h3><p>Tell ChatGPT whether you want a quick rundown or a deep dive. It’s like telling your friend if you’re looking for a quick movie recommendation or a full review.</p><p>Example:</p><pre>Describe the pros and cons of using electric cars in a short paragraph, <br>suitable for a school presentation.</pre><h3>**8. Show the Way: Sample Input:</h3><p>Help ChatGPT understand the conversation’s flow by providing an example of how the chat might start. This gets the ball rolling in the right direction!</p><p>Example:</p><pre>User: Tell me about the benefits of yoga.<br>AI: Yoga offers a range of benefits, from improved flexibility to reduced <br>stress levels.</pre><p>So as a summary: Start with <strong>context</strong>, give <strong>clear instructions</strong>, <strong>choose the response format</strong>, <strong>use follow-ups</strong> for natural flow, <strong>refine </strong>your prompts, <strong>add personality</strong>, <strong>specify length</strong>, and show an example to <strong>set the tone</strong>. With these tips, you’re all set to chat like a pro with ChatGPT!</p><p>Now that you have the tools to craft the perfect prompt, go ahead and have fun exploring the world of ChatGPT! Just remember, it’s like having a virtual buddy — be clear, creative, and curious, and you’ll get amazing responses every time. Happy chatting, and may your prompts be ever perfect!</p><p>Read more about the <a href="https://medium.com/stepstone-tech/technology/home">technologies</a> we use or take an inside look at our <a href="https://medium.com/stepstone-tech/organisation-processes/home">organisation &amp; processes</a>. Interested in working at The Stepstone Group? Check out our <a href="https://medium.com/stepstone-tech/careers-stepstone-d287c9d0d26d">careers page</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c54672a5fc4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stepstone-tech/a-beginners-guide-to-crafting-effective-prompts-for-chatgpt-3c54672a5fc4">A Beginner’s Guide to Crafting Effective Prompts for ChatGPT</a> was originally published in <a href="https://medium.com/stepstone-tech">the-stepstone-group-tech-blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Revealing the secret of a development plan]]></title>
            <link>https://medium.com/stepstone-tech/revealing-the-secret-of-a-development-plan-11a4abdee24f?source=rss----f042839a21a9---4</link>
            <guid isPermaLink="false">https://medium.com/p/11a4abdee24f</guid>
            <category><![CDATA[personal-growth]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[personal-development]]></category>
            <category><![CDATA[company-culture]]></category>
            <category><![CDATA[leadership-development]]></category>
            <dc:creator><![CDATA[Aleksandra Piecuch]]></dc:creator>
            <pubDate>Thu, 03 Aug 2023 11:27:54 GMT</pubDate>
            <atom:updated>2023-08-07T10:14:54.160Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/295/1*ELIw7xpGXFMo-xKNyLiGiQ.jpeg" /><figcaption>Photo by Jan Canty on Unsplash</figcaption></figure><h3>Introduction</h3><p>As a leader, I must admit, I have always dreamed of having a clear development plan throughout my professional career. When I joined The Stepstone Group last year, I was thrilled to discover that this company not only values personal and professional growth but actively supports its employees in this journey. Drawing from my previous experience and the inspiration of this environment, I decided to further explore the topic of development plan while implementing it within my team. In this article, I will share my experience from <strong>The Stepstone Group</strong>, hoping that you can apply some of these concepts to your work, regardless of your role or seniority.</p><h3>Definition of a Development Plan</h3><p>A development plan is a valuable tool for <strong>career growth</strong>, focusing on creating career strategy by both short-term and long-term objectives and the steps required to achieve them.</p><p>It comprises three main pillars: skills assessment, goal setting, and an action plan. My preference is to keep that order during the process. It provides comprehensive overview of personal evaluation, enabling both the leader and the employee collaboratively discuss and strategize career development.</p><p>Through my experience, I’ve observed that many employees are uncertain about the specific direction their careers should take, particularly when there is a high level of seniority within a team. Usually, it means that there are limited or no further options for promotion within the same team or department for them.</p><p>By conducting assessment first, we can ensure that goal setting is based on a solid understanding of the employee’s current capabilities, which can be compared with the employee’s future aspirations and the broader organizational needs, leading to more focused and effective career development.</p><h3>Self-assessment</h3><p>Skills assessment is fundamental to the process as it provides a complete picture of an individual’s strengths, weaknesses, and blind spots. As a leader, your first task is to clearly explain the goals of this assessment, followed by creating a safe space where team members feel comfortable sharing their fears and flaws. <strong>Trust, respect, and kindness</strong> between a leader and team member are crucial for an honest and open self-evaluation.</p><p>To increase self-awareness, the person for whom the development plan is created should fill out a skill gap analysis by themselves. This becomes a valuable starting point for further discussions, especially if there are differences between the employee’s self-assessment and the leader’s evaluation. To assist employees in this process, you can use additional questions <em>(‘What are your weaknesses?’, ‘What are your strengths?’, ‘What do you want to accomplish?’, ‘Which competencies do you still need to develop?</em>’ etc.) or a prepared evaluation form with relevant prompts (see example below).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/443/1*xTefL1eNjvXXo_HkqfPrzA.png" /></figure><h3>Goals</h3><p>Once areas for improvement are identified, the next step is to set goals. These objectives should outline a clear path for personal growth, whether it’s enhancing performance in the current role or transitioning to another position. Goals must adhere to the SMART criteria (Specific, Measurable, Achievable, Relevant, Time-Bound) and be <strong>inspiring.</strong> Collaborating with the employee while setting these goals fosters mutual understanding and aligns expectations.</p><h3>Action plan</h3><p>The final step in building a development plan is creating an action plan. This plan consists of specific tasks to be completed within an agreed timeline, typically a quarter, though the timeframe may vary based on different factors. It’s important that employee can define those key results by themselves. It will increase a sense of <strong>ownership</strong> and the likelihood of success.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/454/1*IWrKL557oTP6QISUiO1CSw.png" /><figcaption>Development plan example</figcaption></figure><h3>Implementing the Development Plan</h3><p>With the development plan in place, personal growth can begin. It’s crucial to provide regular feedback and review progress on tasks and key results to track advancement, identify obstacles, and offer support when needed. To make it <strong>structured and facilitated</strong> toward tracking, I encourage you to follow those questions:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/451/1*4CYQs87GNXPkhB9r10A0ZQ.png" /></figure><p>This journey requires full dedication and commitment from both the leader and the employee.</p><h3>Benefits of Development Plans</h3><p>Based on my team’s experience, having a development plan supported by a regular process yields several advantages. Setting quarterly objectives and reviewing progress monthly allows ample space for <strong>feedback</strong> and discussions regarding both process and personal development.</p><p>Additionally, having a clear direction for growth fosters <strong>motivation</strong> within the team. Utilizing an HR tool to log daily updates, changes, and support for tasks enhances focus, leading to increased achievements and <strong>satisfaction</strong> when goals are accomplished.</p><h3>Continuing the Development Journey</h3><p>A development plan should have a finite timeline but, from a process perspective, it is <strong>infinite</strong>. There will always be opportunities for further development and improvement. If a development plan is successful, take the time to celebrate the achievements. On the other hand, if the plan falls short of expectations, identify the reasons behind it and adjust for the next development planning phase.</p><h3>Tailoring the Plan</h3><p>There is no one-size-fits-all development plan; it should be highly individualized and tailored to each person’s needs and aspirations. When properly implemented, a development plan becomes a powerful tool for increasing motivation and retention among team members. <strong>Embrace</strong> the process and enjoy the <strong>journey of growth</strong>!</p><p>Explore all articles about our <a href="https://medium.com/stepstone-tech/organisation-processes/home">organisation &amp; processes</a>, or read more about the <a href="https://medium.com/stepstone-tech/technology/home">technologies</a> we use. Interested in working at The Stepstone Group? Check out our <a href="https://medium.com/stepstone-tech/careers-stepstone-d287c9d0d26d">careers page</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=11a4abdee24f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stepstone-tech/revealing-the-secret-of-a-development-plan-11a4abdee24f">Revealing the secret of a development plan</a> was originally published in <a href="https://medium.com/stepstone-tech">the-stepstone-group-tech-blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Unveiling Stepstone’s Linguistic Services Superstars: The Heroes Behind the Scenes]]></title>
            <link>https://medium.com/stepstone-tech/unveiling-stepstones-linguistic-services-superstars-the-heroes-behind-the-scenes-16dc25689a19?source=rss----f042839a21a9---4</link>
            <guid isPermaLink="false">https://medium.com/p/16dc25689a19</guid>
            <category><![CDATA[linguistics]]></category>
            <category><![CDATA[ontology]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[stepstone]]></category>
            <category><![CDATA[analytics]]></category>
            <dc:creator><![CDATA[Petra Dlabacova]]></dc:creator>
            <pubDate>Wed, 02 Aug 2023 06:57:59 GMT</pubDate>
            <atom:updated>2023-08-02T07:33:04.913Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UZXqdcdF1oWWwalkXRo0Fg.jpeg" /></figure><p>Join me on an unforgettable journey to the captivating city of Munich, where a fusion of history, remarkable architecture, vibrant people, and, of course, exceptional beer awaits! But beyond its alluring surface, lies a world of linguistic wonders waiting to be explored. Let’s dive into the realm of language superheroes at The Stepstone Group and discover how they are revolutionizing the field of natural language processing and creating linguistic excellence!</p><p>The Stepstone Group’s esteemed Linguistic Services Team (LiSE) comprises a group of linguists, insight analysts and developers spread across Munich, Brussels, Hamburg, and London. During my visit to the Munich Office, I had the privilege of connecting with some of these incredible individuals and gaining valuable insights into their remarkable work. Allow me to share my impressions.</p><p>Superhero Synergy in Action: A Day in the Life of Linguists, Developers, and Insight Analysts — the Linguistic Services Team</p><p>So, what does a typical day in the life of our Linguistic Services Team entail? Picture a cohesive team of linguists, software developers, and insight analysts working seamlessly together. They kick-start their day with daily LiSE stand-up meetings, where they share updates, discuss vital topics, and identify opportunities for process optimization. These quick catch-up sessions have become the foundation for seamless communication and collaboration. They engage in discussions about ongoing projects, address challenges, and optimize processes to ensure the highest quality of service.</p><p>Our Linguistic Services team is also involved in various projects, such as developing infrastructure, ensuring data accuracy, and validating internal standards and external resources. Their attention to detail, language expertise, and adaptability to evolving requirements make them indispensable members of our Tech community. The linguists collaborate closely with developers to ensure the highest quality of our software.</p><p>And what do the insight analysts do? The job of the insight analysts in the team is to derive insights from The Stepstone Group’s data. They meticulously analyse all kinds of data and identify meaningful patterns and trends. These invaluable insights form the basis of numerous ad-hoc analyses that the backend teams utilize to evaluate and optimise their products.</p><p>However, the contributions of our esteemed Linguistic Services team extend even further. Proudly donning the moniker “human eye on AI”, they diligently test and scrutinize the work of other teams. With their discerning eyes, they ensure that our AI-powered systems maintain the highest standards of accuracy and precision. Their efforts have earned them well-deserved recognition as a crucial element of our operations.</p><p>Empowering Communication with Natural Language Processing</p><p>During my brief yet awe-inspiring stay, I dove deep into the realm of natural language processing (NLP), powered by ontologies and dictionaries. The Stepstone Group’s team of linguistic wizards operates on a global scale, ensuring linguistic precision and cultural respect across their platform.</p><p>Their ontologies play a pivotal role in facilitating seamless job searches, automated resume screenings, and even the creation of witty chat widgets. Through their expertise, they empower users to connect with their dream opportunities.</p><p>Cracking the Code of Job Postings</p><p>Imagine linguistic detectives, armed with a mastery of words akin to the renowned Sherlock Holmes. The Stepstone Group tools strategically analyse job postings to extract crucial details, including skills, qualifications, professions, and sectors. Through this meticulous work, job seekers can find the right job. Armed with their linguistic wizardry, the linguists decode the intricate language of job descriptions facilitating meaningful connections between job seekers and employers.</p><p>Elevating User Experience through Linguistic Excellence</p><p>The Stepstone Group’s language assets extend beyond the realm of language processing, playing a vital role in user experience research and testing. Acting as Lingua Franca mediators, they optimize the platform’s linguistic elements to enhance user satisfaction.</p><p>In addition, our superheroes also provide invaluable support in resolving customer issues encountered while using our system. Ensuring every concern is promptly addressed and resolved to the utmost satisfaction of our customers. They recognize the significance of delivering exceptional service.</p><p>The Power of Databases</p><p>To support their linguistic endeavours, our team relies on two powerful databases.</p><p>TexLex serves as a central hub, meticulously collecting and organizing words, concepts, and data to facilitate seamless operations. With its sophisticated semantic structure, the database enables numerous teams to search, retrieve, and analyse vast amounts of data effortlessly. It houses essential information, such as educational levels, benefits, job roles and titles, and work policies, among others. Serving as our primary languages — English, German, Dutch, French and Hebrew — TexLex ontology acts as a “common language” across The Stepstone Group, ensuring standardized communication and a vast network of interconnected multilingual data.</p><p>The Stepstone Group’s linguistic team is continuously expanding and updating their databases, creating new roles and pronouns, and providing support to other teams. Their collaborative efforts result in an impressive number of skills and job descriptors (JDs) added to their databases. In fact, in June alone, they added 16,000 skills, bringing the total to almost 237,000 skills in 11 languages. On the other hand, GeoLex, our geo database, prioritizes location matching, offering valuable geographic information to enhance search capabilities. It encompasses a comprehensive hierarchy of location names, translations, latitude and longitude coordinates, and post codes (area codes). Managed by LiSe, GeoLex comprises around 550,000 locations organized from the world level down to countries, cities, and micro-locations.</p><p>With the powerful combination of their TexLex and GeoLex databases, they unlock untapped potential in the realm of linguistic possibilities.</p><p>Collaboration and Continuous Improvement</p><p>At The Stepstone Group, we believe in the power of collaboration to achieve greatness, particularly when it comes to language. Our Linguistic Services team comprising of linguists, developers, and insight analysts, operates like a well-oiled machine. Working hand in hand with various departments, they establish comprehensive language usage guidelines, ensuring our software reaches new heights of quality and be able to say: <a href="https://medium.com/stepstone-tech/our-journey-how-to-become-a-value-driven-company-9f94a07f06ae">we stand on our number</a>.</p><p>To support our linguistic endeavours, we have dedicated front-end and back-end developers. They work seamlessly to craft clean and efficient code that powers the creation and maintenance of the ontology. While front-end development focuses on creating intuitive interfaces and delivering a seamless user experience, the back-end developers tackle complex tasks, such as data processing, feature integration, infrastructure improvements, APIs and exports. Their efforts ensure that other teams can access the wealth of linguistic information easily.</p><p>Together, our developers play a crucial role in shaping and enhancing the foundation of our linguistic services. From implementing tag groups that bring order to the chaos, to crafting sorting groups that simplify your life, and seamlessly integrating user interfaces that leave you in awe, they’ve got it all in the works.</p><p>Of course, none of this would be useful without our experienced and perceptive insight analysts who play a vital role by extracting valuable insights, evaluating business strategies, and providing recommendations for optimization. Armed with analytical skills and supported by linguistic know-how, they become instrumental in fine-tuning our user experience, improving search results, and make our language-driven features and functionalities sing. They ensure that our decisions are grounded in solid evidence, leading to happier users and perfectly aligned business goals. The collaborative efforts of our insight analysts significantly contribute to the continued success and growth of our linguistic services.</p><p>Furthermore, the data insights are presented through informative dashboards accessible to a wider audience. Whether it’s B2B or B2C topics, there is a dedicated dashboard for each use case, containing all the required information for detecting irregularities and monitoring key metrics. With their work the insight analysts play a pivotal role in helping the business make informed decisions and continually enhance overall performance.</p><p>Conclusion</p><p>My encounter with our linguistic superheroes at The Stepstone Group in Munich was nothing short of extraordinary. Witnessing their transformation of linguistic complexity into a harmonious symphony was truly awe-inspiring. As they revolutionize the business world, one word at a time, stay tuned for more adventures on their quest to create unparalleled linguistic excellence. The Stepstone Group’s language superheroes are changing the game, and you don’t want to miss it!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/686/1*Twuwxp5QNCDJw5wK8tVyJg.jpeg" /></figure><p>Read more about <a href="https://medium.com/stepstone-tech/about-thestepstonegroup/home">The Stepstone Group</a>, the<a href="https://medium.com/stepstone-tech/technology/home"> technologies </a>we use or take an inside look at our <a href="https://medium.com/stepstone-tech/organisation-processes/home">organisation &amp; processes</a>. Interested in joining us? Check out our <a href="https://medium.com/stepstone-tech/careers-stepstone-d287c9d0d26d">careers page</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=16dc25689a19" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stepstone-tech/unveiling-stepstones-linguistic-services-superstars-the-heroes-behind-the-scenes-16dc25689a19">Unveiling Stepstone’s Linguistic Services Superstars: The Heroes Behind the Scenes</a> was originally published in <a href="https://medium.com/stepstone-tech">the-stepstone-group-tech-blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Rama LLAMA Ding Dong — an overview of four key LLMs: GPT, Llama, Claude and Bard]]></title>
            <link>https://medium.com/stepstone-tech/rama-llama-ding-dong-an-overview-of-four-key-llms-gpt-llama-claude-and-bard-e78681e47416?source=rss----f042839a21a9---4</link>
            <guid isPermaLink="false">https://medium.com/p/e78681e47416</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[stepstone]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[bard-ai]]></category>
            <dc:creator><![CDATA[Thiemo Bubel]]></dc:creator>
            <pubDate>Mon, 24 Jul 2023 08:36:03 GMT</pubDate>
            <atom:updated>2023-07-24T08:43:48.347Z</atom:updated>
            <content:encoded><![CDATA[<h3>Rama LLAMA Ding Dong — an overview of four key LLMs: GPT, Llama, Claude and Bard</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*eYxXoyklk9PfQ4iq" /><figcaption>Photo by <a href="https://unsplash.com/@paulequay?utm_source=medium&amp;utm_medium=referral">Paul Lequay</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The launch of ChatGPT last November made a big splash. More than half a year later it is no longer the only game in town and many alternative large language models (LLMs) with slightly different strengths and weaknesses have been released. An LLM is a trained deep-learning model that understands and generates text in a human-like fashion. Just this week, Meta launched its Llama2 and made it widely available.</p><p>Here’s a high-level summary of four key LLMs, including their associated advantages and challenges.</p><h3>OpenAI / GPT 4.0 (Microsoft)</h3><ul><li>Advanced language understanding and generation capabilities for a wide range of topics.</li><li>Impressive coding ability and sophisticated word choice</li><li>Suitable for content creation and coding tasks.</li><li>However, criticized for lack of transparency in model training process and the model is only available by using one of OpenAI’s services like user interfaces or APIs.</li></ul><h3>Llama 2.0 (Meta)</h3><ul><li>Exceptional performance in generating helpful responses for single and multi-turn prompts.</li><li>Suitable for chat applications and tasks requiring human interaction.</li><li>However, comparatively lower coding capability.</li><li>Open-source nature allows the model to be run in a company’s environment and larger control to finetune it to the specific domain and use cases.</li></ul><h3>Claude 2.0 (Anthropic)</h3><ul><li>Excellent at tasks involving coding, mathematics, logical thinking, and document understanding.</li><li>Proficient in processing large documents including PDFs, a challenging task for other models that are more limited regarding the maximum input size.</li><li>Impressive Python coding skills, scoring 71.2% on Codex HumanEval.</li><li>Similar to GPT 4.0, accessible via an API but not fully open source.</li></ul><h3>Bard (Google)</h3><ul><li>Designed for conversational AI applications, capable of generating human-like responses and providing advice, answering questions, and generating content.</li><li>Draws on information from the web to provide fresh, high-quality responses.</li><li>Limited in comparison to GPT-4, particularly in providing specific and accurate responses for certain tasks like CSS coding.</li><li>Currently not recommended to use at StSt because it doesn’t allow to block data sharing with Google.</li></ul><h3>Sources:</h3><ul><li><a href="https://medium.com/@sanjeeva.bora/the-new-era-of-generative-ai-unveiling-gpt-4-0-llama-2-0-and-claude-2-0-932b026ecfe7">The New Era of Generative AI: Unveiling GPT 4.0, Llama 2.0 and Claude 2.0</a></li><li><a href="https://analyticsindiamag.com/llama-2-vs-gpt-4-vs-claude-2/">Llama 2 vs GPT-4 vs Claude-2</a></li></ul><p><em>Co-authored by Felicitas von Rauch</em>, <em>Timm Lochmann and Heinke Hihn</em></p><p>Read more about <a href="https://medium.com/stepstone-tech/about-thestepstonegroup/home">The Stepstone Group</a>, the<a href="https://medium.com/stepstone-tech/technology/home"> technologies </a>we use or take an inside look at our <a href="https://medium.com/stepstone-tech/organisation-processes/home">organisation &amp; processes</a>. Interested in joining us? Check out our <a href="https://medium.com/stepstone-tech/careers-stepstone-d287c9d0d26d">careers page</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e78681e47416" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stepstone-tech/rama-llama-ding-dong-an-overview-of-four-key-llms-gpt-llama-claude-and-bard-e78681e47416">Rama LLAMA Ding Dong — an overview of four key LLMs: GPT, Llama, Claude and Bard</a> was originally published in <a href="https://medium.com/stepstone-tech">the-stepstone-group-tech-blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>