<?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[Stories by Parth Vasa on Medium]]></title>
        <description><![CDATA[Stories by Parth Vasa on Medium]]></description>
        <link>https://medium.com/@parth.vasa?source=rss-7ebff91e275a------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*n7irrtfeNv6GaIDZ.</url>
            <title>Stories by Parth Vasa on Medium</title>
            <link>https://medium.com/@parth.vasa?source=rss-7ebff91e275a------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 07:54:35 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@parth.vasa/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com" rel="hub"/>
        <item>
            <title><![CDATA[Sorting By Time]]></title>
            <link>https://medium.com/@parth.vasa/sorting-by-time-efa5e04ec2a0?source=rss-7ebff91e275a------2</link>
            <guid isPermaLink="false">https://medium.com/p/efa5e04ec2a0</guid>
            <category><![CDATA[search]]></category>
            <category><![CDATA[search-engines]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[recommendations]]></category>
            <category><![CDATA[recommendation-system]]></category>
            <dc:creator><![CDATA[Parth Vasa]]></dc:creator>
            <pubDate>Tue, 07 Feb 2023 16:02:44 GMT</pubDate>
            <atom:updated>2023-02-09T14:05:32.905Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Ni1hThqwpKAM8ZOr" /><figcaption>Photo by <a href="https://unsplash.com/es/@morganhousel?utm_source=medium&amp;utm_medium=referral">Morgan Housel</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Sorting by time, aka<em> reverse chronological sort</em> aka <em>chrono-sort </em>should have never been a big deal. But it is. For a long time, if you were not going to sort something alphabetically you’d sort it by time of creation or update. Then, searchable universe grew and with search engines came relevance-ranking. Social media feeds took that to a whole new level with their near infinite content and relevance-ranking. Chrono-sort became this cool vintage thing that people pined for.</p><p>Relevance ranking is the right choice for vast majority of our needs. Still, social media nostalgia aside, there are some clear use cases where sorting by time is important. That’s what this note is about.</p><p>Fundamentally, people want time-sorting over relevance-ranking <em>when recall is much more valued than precision</em>. Since normal people don’t think in terms of precision/recall, let’s look at some examples needs that make people crave more recall.</p><p>In Search, the need/intent for time-sort is pretty explicit</p><ol><li><strong>When search criteria implicitly is time-related</strong> ( “I remember this happened when last month when that thing was going on”). This is why the <em>majority of email searches are time sorted</em>.</li><li><strong>When time is the safest search criteria :</strong> This happens when efficiency of finding is less important than completeness of coverage — meaning the cost of missing a single possible result is higher than cost of time wasted in looking through results (100% recall). This is why<em> compliance, diligence or legal search are often time-sorted.</em> Time matters less here but the fear that ranking by anything else might make you miss a result, is the driving force.</li></ol><p>For Recommendation products, on the other hand, the intent for time-sorting is more subtle. It’s either FOMO or its more intense cousin : loss of trust.</p><ol><li>Fomo is when a person doesn’t want to miss something important that may have happened — your crush posted a picture but you missed it, or a limited edition sneaker dropped but you didn’t catch it fast enough.</li><li>Loss of trust is the same event but you assume bad intent by the system. You missed your crush’s picture <em>because</em> the ranking algo drowned you in influencer videos, or you missed the sneaker sale <em>because</em> the phone decided to push other reminders that you interact with more often.This gets built over time.</li></ol><p>So how do you help users with these?</p><p>You can, of course, make your ranking so sophisticated, use time and freshness in your ranking so wisely, that people would never face any of these problems. Maybe that’s possible. But since we are dealing with people and everything above is more about feelings than facts, chances are they still will face some of these problems.</p><p>Here are some patterns at the product level that I have seen work well.</p><p>In search the time-sorting intent is pretty explicit — they demand it — so the patterns should be based around helping the users make sense out of the results.</p><p><strong>Add value beyond ranking results </strong>: You may no longer be responsible for up-ranking the most relevant result but you are still responsible for helping them find the right one and optimize their time.</p><p>So you can use more your sophisticated ML / Search techniques to help the user beyond search ranking. This can be</p><ol><li>Improved retrieval like query expansion, document understanding, so that you are casting a smarter/wider net even though you are sorting results by time.</li><li>Use content-understanding to show better metadata about each result that allows them to quickly reject results that are not relevant.</li><li>Use auto-suggest, query-refomulation etc to help them form better search queries. Google has trained us to not use things beyond nouns and operators, but we can be reprogrammed with success.</li></ol><p><strong>Show the best answer and then time-sort everything else :</strong> In this case you would effectively run two searches — one with a more sophisticated search/reasoning system and one in a pure retrieval system. And in your results screen, you would show them the best answer as a hero-result (or maybe show top 3). And then show them everything in chrono order.</p><p>In Recommendations products, time-sorting intent is implicit so the patterns are more about context and support to gain trust and give control.</p><p><strong>Contextual hints to reduce FOMO :</strong> This can come in different forms but they are usually subtle. It can be a hint when you are done with fresh content or when there is fresh content you haven’t seen; explanation of why some old content was up-ranked; hint when some account you interact with has fresh content or context around when they last shared. Etc</p><p>If you use contextual hints, be sure to make them non-intrusive and test them a thoroughly (both qual and quant). It’s very easy to go from being helpful to being a pest.</p><p><strong>Grouped sorting</strong> : Instead of having one flat list, group the results and then sort. There are generally two kinds of groupings</p><ol><li>Grouping by a <em>theme</em> and then time-sorting within that theme. The themes can be then ranked by relevance. <em>Notifications ranking </em>is a good example for this : cluster by app, show notifications in time order in that cluster, rank clusters by relevance.</li><li>Grouping results by <em>time</em> and relevance ranking them. E.g. Today’s news, last week’s news and older news (all sorted by relevance inside of those timeframes)</li></ol><p><strong>Easy filtering and searching :</strong> Show them a ranked list but allow them to easily filter by source (e.g. account or app) or time or topic. I have seen this happen in a few different ways like allwoing a contextual search or Up-leveling “Facets” as clickables (source, topic, ratings, some other derived information</p><p>Hope this was interesting. What did I miss?</p><p>Note : I wrote a version of this note for my team back in 2016 at Bloomberg but never published it, then again in 2018 at Instagram (again never published it). Finally, last weekend, when New York was painfully cold, I decided to go back and finish it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=efa5e04ec2a0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[About designing orgs]]></title>
            <link>https://medium.com/@parth.vasa/about-designing-orgs-7fe64c33f7cb?source=rss-7ebff91e275a------2</link>
            <guid isPermaLink="false">https://medium.com/p/7fe64c33f7cb</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[engineering-mangement]]></category>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[organizational-culture]]></category>
            <category><![CDATA[org-design]]></category>
            <dc:creator><![CDATA[Parth Vasa]]></dc:creator>
            <pubDate>Mon, 19 Sep 2022 19:44:56 GMT</pubDate>
            <atom:updated>2022-09-19T21:43:50.708Z</atom:updated>
            <content:encoded><![CDATA[<h3>About Designing Orgs</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*73dechvhBKVUZsOS" /><figcaption>Photo by <a href="https://unsplash.com/@lidyanada?utm_source=medium&amp;utm_medium=referral">Lidya Nada</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>This is a note about designing organizations. Not really about how to design/redesign them — but about my beliefs on why we design them.</p><p>As leaders (especially engineering leaders) org design is probably the most leveraged thing we can do. Since it’s the act of dividing our group of people into teams and then connecting those teams together, it’s worthwhile getting into why we design teams in the first place.</p><p><strong>Note : </strong>Most discussion here is very engineering focussed but the principles should apply to any org.</p><h4><strong>What do you design for?</strong></h4><p>My strong belief is that the objective of any org design should always be <em>Execution Efficiency and Quality</em>. All other optimizations like career growth of strong people, culture building etc are secondary (albeit important) objectives and should be applied after an ideal org design is made without them in mind.</p><p>There are three fundamental factors in how we design orgs ( meaning break a group of people into teams) and they interplay and trade-off while building orgs.</p><ol><li>Size : After a certain size a flat group of people starts to become inefficient. This is a self explanatory but really important constraint.</li><li>Community / Skills : People benefit from working closely with others with similar skill sets and varying experiences. It gives people a chance to mentor, be mentored and grow. <em>The quality of solutions is better when an org optimizes for skills growth.</em></li><li>Identity / “Charter” : The product benefits when a smart group of talented people live and breathe a particular problem end-to-end . <em>The quality of problems solved is better when an org optimizes for ownership of charters.</em></li></ol><h4><strong>Horizontal or Vertical</strong></h4><p>Take for example the classic Horizontal vs Vertical Engineering org debate. It’s basically a community vs identity tradeoff.</p><p><strong>Optimizing for community creates horizontal orgs</strong> : This optimizes for liquidity and development of talent. People operate and grow faster if they work in a community of others like them. For example, ML engineers would function better in a team full of other ML engineers vs being fragmented with a group of Mobile developers.This advocates for <em>horizontal</em>, skill set based teams (e.g. Mobile, Web, ML, Data engineering …).</p><ul><li>This kind of design helps you hire and onboard engineers faster. In larger orgs it lets you hire more senior engineers who are domain experts in those domains and can influence a large team’s work. It also creates a better tech stack and better solutions.</li><li>But you run the risk of the team’s <em>identity</em> being too tied up with a solution vs the problem. <strong>The “how” starts to rule and the “why” gets lost, creating a loss of empathy towards the product.</strong> Engineers start to feel that end-to-end experiences are someone’s else’s (usually the PM’s) problem. Since there will always be more engineers than most other functions, you have the majority of the org not thinking about the end product but only their part of the solution.</li></ul><p><strong>Optimizing for Identity creates vertical orgs</strong> : This optimizes for clear charters and end to end ownership. The quality of product is better if a team is accountable for a problem end to end. This advocates for <em>vertical, </em>“problem” based, teams (e.g. Onboarding, Engagement, Growth, Content Creation, Trading…).</p><ul><li>This creates strong empathy in the engineers for the end product vs a piece of a solution. <strong>The “Why” and the “what” become the first class citizens. </strong>It also creates a strong customer focus in engineering and as a result a big part of the org. This typically creates better products.</li><li>On the other hand, this can create duplication of technical solutions for similar problems by groups that operate in silos. If not done carefully the consistency of the codebase starts to deteriorate and eventually that inconsistency creeps into the product. You start shipping your org chart.</li></ul><p>So what’s the right optimization? Neither and both. Based on the org’s lifestage and size you usually go from one to the other and then both — back and forth.</p><p><strong>Let’s take an example</strong></p><p>You start a company (or a new org). It’s all just one small team that has all kinds of people you need (or the people you have to do all the work you need).This is the most vertical team you will ever build. Everyone has the same identity, work gets done fast and there is very little coordination needed.</p><p><strong>Scale Stage 1</strong></p><p>As the team grows you start to add more people with some skill sets so that you can produce more work faster. You are still working on mostly the same problem/project but with more people. It makes sense to start to organize horizontally based on skills at first. Like Mobile team, backend team, ML team etc. Why?</p><ul><li>At this stage a strong community will create the most amount of efficiency in terms of people swarming on important work. It also helps you onboard new people the fastest.</li><li>It allows you to make foundational decisions about the technical architecture.</li><li>The identity already is pretty clear at this stage as you are focussed on a small number of problems and the company/org leadership is able to inculcate it at this size.</li></ul><p><strong>Scale Stage 2</strong></p><p>But as you grow, the problem space starts to grow. You start to have multiple top-level problems that you are chasing. Your teams will start to get split into multiple “pods”.</p><p>For example, your mobile team is now divided into 3 product projects pods plus a platform team. They work with a backend team which has a somewhat similar split. You start to see non-engineering functions like PM/Design start to be mapped to these “ pods” at first for a project but slowly becoming permanent. In essence these functions act as an adapter for a world that has become verticalized over an org that is still horizontal.</p><p>This is a fine short term solution but as the team keeps growing and pods become semi-permanent, you should consider a structural change. Why?</p><ul><li>Identity starts to become really important at this point because individual problems become big enough to need end to end thinking. And that e2e thinking can’t just be done by PMs or leadership — you want the whole team owning a problem.</li><li>Decision making becomes easier and more scalable and metrics and success is easier to measure</li><li>The team sizes may be big enough by now that even the vertical teams will be able to provide “community” in terms of skill growth. For example, if you were to divide by user objectives (say Engagement, Trading, Community growth… ) each of these parts might have enough Mobile/ Backend / ML engineers to provide community and mentorship.</li></ul><p><strong>Scale stage 2.5 and so on</strong></p><p>You can’t verticalize everything. Some parts would need to stay horizontal. And some parts may need to be cut out of a vertical team and made horizontal. Typically, these are parts which can have their own product identity even as a horizontal.</p><p>In my experience it’s one of the two things</p><ul><li>Code/functionality that is (or should be) shared and has 2 or more users — aka a platform.</li><li>A competency that is so core to the org/company that it has to stay “skill-based” and be optimized independently. For example, image understanding, or user modeling or OS kernel.</li></ul><p>Conversely, you may have something laid out as a horizontal that should merge with another team to form a vertical team. Maybe there is a platform team that only serves one client team (or mostly one client team). It was made horizontal to give it a community but it might be time to bring it closer to its most demanding users.</p><p>The question always is when do you go from one stage to the other. Let’s chat about all that next time. I am surprised you made it this far!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7fe64c33f7cb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your first 6 weeks]]></title>
            <link>https://medium.com/@parth.vasa/your-first-6-weeks-as-a-new-engineering-leader-b0badb781866?source=rss-7ebff91e275a------2</link>
            <guid isPermaLink="false">https://medium.com/p/b0badb781866</guid>
            <category><![CDATA[engineering-mangement]]></category>
            <category><![CDATA[first-30-days]]></category>
            <category><![CDATA[first-90-days]]></category>
            <category><![CDATA[getting-started]]></category>
            <dc:creator><![CDATA[Parth Vasa]]></dc:creator>
            <pubDate>Sun, 10 Apr 2022 20:29:19 GMT</pubDate>
            <atom:updated>2022-05-20T01:46:55.721Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*B7AkD6VrM1bui1An" /><figcaption>Photo by <a href="https://unsplash.com/@cliqueimages?utm_source=medium&amp;utm_medium=referral">Clique Images</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>This is a note about getting started as an experienced leader in a new role. You already know how to do your job. Your ramp up is not about re-learning it but about learning to do it in your new team. So, take time to build trust and context for driving impact in the long term vs trying to drive big changes in the short term. This will take time and that’s ok.</p><p>Your first six weeks are going to be all about learning. To make it more concrete, think of these as deliverables for your ramp up.</p><ul><li>First deliverable is to build Relationships that will allow you to learn</li><li>Then use those to build POV about your team, product and organization.</li><li>Third deliverable is to identify small value adds and execute on them to build trust. Think Ted Lasso and water pressure.</li><li>Then, identify larger value adds.</li><li>Repeat</li></ul><p>In some ways this is similar to the classic Understand → Identify → Execute framework applied to org leadership. So you understand an area, then identify opportunities of making things better, then execute on them, while understanding another area<strong>. </strong>Always in that order, always on repeat, not always serially .</p><p>In my experience, learning priorities in early days are divided into four areas. The order of them may change based on your role, your team, whether you had a predecessor or not etc but it will usually come down to these</p><ol><li>Team Health</li><li>Team Capacity and Performance</li><li>Product Strategy and execution</li><li>Leadership connection and trust</li></ol><p>The best way to learn is,of course, asking questions. So meet as many people as you can and ask a lot of questions. Here the list of some questions I have used.</p><h4><strong>Team Health</strong></h4><ul><li>How are people feeling?</li><li>Do people trust each other?</li><li>What is in their way ?</li><li>Is there a sense of career progression?</li><li>What are some big things that are missing ?</li><li>What are some small things that are missing?</li><li>Who are the leaders?</li><li>Who are the culture careers?</li><li><em>Identify : Hotspot and strong parts of the team health</em></li><li><em>Implement : Some small trust wins</em></li></ul><h4>Hiring</h4><ul><li>How is our hiring going?</li><li>Which teams are staffed well and which not?</li><li>How is the attrition ?</li><li>Do managers have a good hiring pitch?What is their sell process? Is everyone selling the same thing?</li><li>Do managers know why hiring is tough ( if so)?</li><li>How many do we source via referrals? What would make people refer more?</li><li>Retrospect on people that turned us down</li><li><em>Identify : Some areas to make the pipelines go faster.</em></li><li><em>Implement : Small optimizations</em></li></ul><h4>Product/Business Strategy</h4><ul><li>What is our northstar?Why is that important?</li><li>How do we measure it?</li><li>What has PMF? What doesn’t and why</li><li>What are the bottlenecks / barriers to growth ?</li><li>Who are our outside partners and how is the relationship?</li><li><em>Deliverable : being able to talk intelligently about the product, plans, metrics and goals at the end of 6 weeks.</em></li><li><em>Identify : Some opportunities to unblock/speed up the execution and have everyone talk about the product in the same way</em></li></ul><h4>Org Design</h4><ul><li>Why is it org designed out as it is right now?</li><li>What are the concerns with it? What is working well?</li><li>How should the org be laid out?</li><li>Do we have the right leadership for the right design?</li><li><em>Identify: POV on what if anything needs to change in the org of how we support it</em></li></ul><h4>Product development</h4><ul><li>Roadmapping / sprint planning</li><li>How much do engineers get involved and at what stage?</li><li>Are we shipping what we build ? If not, why not?</li><li>Is decision making clear?</li><li>How much time do we spend in new features vs bugfixing vs platform work</li><li><em>Identify : Can we prototype and test faster?</em></li></ul><h4>Development Processes</h4><ul><li>How are dashboards / alarms</li><li>How is engineering Quality tracked?</li><li>How is the oncall process — shadow an oncall</li><li>How hard is it to develop a small feature</li><li>How are SEVs handled?</li><li>Are there retrospectives?</li><li>Identify : Some development pain points</li><li>Execute : Some quick wins or unblocks</li></ul><h4>Leadership and IC connection</h4><ul><li>Do they know their leaders? Trust them?</li><li>Do they know who makes which decisions?</li><li>Is there a regular forum for them to listen/talk to their leaders?</li><li>Is there a process to get people’s ideas in the roadmap?</li><li>Identify : One misconception you can fix or gap you can fill</li></ul><p>Once you have done enough meetings and Q&amp;A sessions make a list of priorities that you want to work on next. Share that with your peers and/or your manager and see how they feel about them. Whether or not they agree with them or are aware of them would give you an important signal about the work that’s cut out for you.</p><p>Hope this helps. Next time, maybe we talk about trust building or org building.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b0badb781866" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>