<?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 Shruti Pandey on Medium]]></title>
        <description><![CDATA[Stories by Shruti Pandey on Medium]]></description>
        <link>https://medium.com/@srtpan?source=rss-47a91e766c89------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ii9N0PrWua0nRgvpNhT1Jw.jpeg</url>
            <title>Stories by Shruti Pandey on Medium</title>
            <link>https://medium.com/@srtpan?source=rss-47a91e766c89------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 14:17:29 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@srtpan/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[Getting Started with LangChain]]></title>
            <link>https://srtpan.medium.com/getting-started-with-langchain-1441215e5673?source=rss-47a91e766c89------2</link>
            <guid isPermaLink="false">https://medium.com/p/1441215e5673</guid>
            <category><![CDATA[llm-applications]]></category>
            <category><![CDATA[langchain]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[chatbots]]></category>
            <dc:creator><![CDATA[Shruti Pandey]]></dc:creator>
            <pubDate>Sat, 26 Jul 2025 06:45:43 GMT</pubDate>
            <atom:updated>2025-07-26T06:47:19.219Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xpaVUwfxVhEDVpcjn_1eUA.avif" /><figcaption>Lanchain Logo and Text</figcaption></figure><p>Large Language Models (LLMs), such as ChatGPT, Claude, etc., have become an integral part of our everyday lives. Every time I am commuting, I see at least one person chatting with ChatGPT or Claude, which shows how pervasive it has become. As an AI developer with a deep curiosity to develop LLM-powered apps utilizing foundation models, you might have come across LangChain. If you’d like to get started and get a primer on how to use it, this guide is for you. This gets you started with LangChain with a hands-on exercise towards the end to set you up for building larger projects with LangChain.</p><h4>What is LangChain?</h4><p>LangChain is a powerful open-source framework that simplifies building LLM-powered applications with various building blocks and abstraction layers. The framework provides standardized interfaces for working with different LLMs, memory management, prompt engineering, and chain composition.</p><p>What makes LangChain so versatile and essential in LLM application development are these capabilities:</p><ol><li><strong>Various Model Providers</strong>: You can switch between LLMs of your choice inside LangChain with a choice of GPT, Anthropic, Gemini, Vertex, Azure, and other models. Depending on your task and use case, you have the ability to pick your model inside one environment.</li><li><strong>Memory Management:</strong> It enables you to maintain conversation history, from simple buffer memory to more sophisticated summarization approaches. If you are more interested in Retrieval-Augmented Generation (RAG), LangChain has a ton of options to connect with various Vector Databases like Chroma DB, Pinecone, AstraDB, etc.</li><li><strong>Prompt Management: </strong>This is a very useful feature inside LangChain that enables you to perform prompt engineering to get the required outputs with your model. There are many ChatPromptTemplates available for use inside LangChain that you can access inside the documentation: <a href="https://python.langchain.com/api_reference/core/prompts.html">https://python.langchain.com/api_reference/core/prompts.html</a>.</li><li><strong>Chain Composition:</strong> Another very useful feature. It allows you to connect various processing steps together in a sequence, so the output of one step becomes the input to the next step. In the hands-on practice, I will show an example of chaining and how you can use it.</li></ol><h4>LangGraph In Brief</h4><p>Complex applications that require more sophisticated orchestration are a suitable candidate for LangGraph. It provides a graph-based approach to build stateful, multi-step applications where different nodes can represent various operations like API calls, database queries, or different LLM interactions, particularly helpful for making agents.</p><h4>LangSmith In Brief</h4><p>LangSmith is a companion platform of LangChain and provides options to track token usage, latency, monitor outputs, A/B test prompts, model configurations, and so much more. Basically, a comprehensive observability platform for your LLM application.</p><h3>Let’s get building: Lipid Panel Chatbot</h3><p>Let’s explore these concepts with some hands-on practice. I will walk over a simple implementation of a lipid panel support chatbot using LangChain with Anthropic’s Claude model.</p><pre>#start by installing the langchain anthropic library. <br>#They have different libraries for other foundation models - I am using Anthropic<br><br>!pip install langchain-anthropic<br>!pip install langchain<br>!langchain-community<br>!pip install dotenv #to manage environment variables</pre><p>The next step is to configure the API key for the foundation model you will be using. I am using Claude. You can get the API_KEY by logging in to the Anthropic console, creating an account, adding some credits, and adding an ANTHROPIC_API_KEY. Make sure to save the key in a safe spot. If you are following the tutorial in a notebook, you can use the following code, or you can save the key as an environment variable if you wish you deploy it as an application.</p><pre>import getpass<br>import os<br><br>if not os.environ.get(&quot;ANTHROPIC_API_KEY&quot;):<br>  os.environ[&quot;ANTHROPIC_API_KEY&quot;] = getpass.getpass(&quot;Enter API key for Anthropic: &quot;)<br><br>from langchain.chat_models import init_chat_model<br><br>llm = init_chat_model(&quot;claude-3-5-sonnet-20241022&quot;, model_provider=&quot;anthropic&quot;)</pre><h4>Messages inside LangChain</h4><p>There are some types of messages that you can use (access the full guide here: <a href="https://python.langchain.com/docs/concepts/messages/">https://python.langchain.com/docs/concepts/messages/</a>). From the official guide: <em>“Messages are the unit of communication in </em><a href="https://python.langchain.com/docs/concepts/chat_models/"><em>chat models</em></a><em>. They are used to represent the input and output of a chat model, as well as any additional context or metadata that may be associated with a conversation”</em>. These are the different types of messages:</p><ul><li><strong>HumanMessage: </strong>A message from the user</li><li><strong>AIMessage:</strong> A message from the LLM</li><li><strong>SystemMessage:</strong> A message that tells the language model how to behave</li></ul><h4>ChatPromptTemplate</h4><p>Using LangChain’s ChatPromptTemplate, you can create structured prompts that separate system messages from user messages. This ability makes the responses more aligned, makes the code more maintainable, and allows for easy prompt iterations.</p><p>For specialized information like lipid panel or your internal documentations, you probably need information outside LLMs. The chatbot leverages context to provide relevant medical information about lipid panels, interpreting test results, and offering general guidance. The ChatPromptTemplate helps with consistent formatting while allowing dynamic content insertion based on user queries.</p><p>Let’s see it in action.</p><pre>    import dotenv<br>from langchain.prompts import (<br>    PromptTemplate,<br>    SystemMessagePromptTemplate,<br>    HumanMessagePromptTemplate,<br>    ChatPromptTemplate,<br>)<br><br>dotenv.load_dotenv()<br><br># prompt_template contains the instructions that you&#39;ll pass to the model, delimited with context and question<br><br>prompt_template = &quot;&quot;&quot;Your job is to answer questions about their lipid panel. <br>The user would have received an evaluation of their lipid profile <br>and would like to understand the report in simple terms and <br>how they can improve. Use the provided context to answer questions. <br>Be as detailedas possible, but don&#39;t make up any information that&#39;s not<br>from the context. If you don&#39;t know an answer, politely decline that you <br>don&#39;t know the answer<br>{context}<br>{question}&quot;&quot;&quot;<br><br>#making use of system prompt and human prompt templates <br>system_prompt = SystemMessagePromptTemplate(<br>    prompt=PromptTemplate(<br>        input_variables=[&quot;context&quot;],<br>        template=prompt_template,<br>    )<br>)<br><br>human_prompt = HumanMessagePromptTemplate(<br>    prompt=PromptTemplate(<br>        input_variables=[&quot;question&quot;],<br>        template=&quot;{question}&quot;,<br>    )<br>)<br>messages = [system_prompt, human_prompt]<br><br>prompt_template = ChatPromptTemplate(<br>    input_variables=[&quot;context&quot;, &quot;question&quot;],<br>    messages=messages,<br>)<br><br>#using the chain in LangChain to combine the chatprompttemplate and LLM<br># LangChain Expression Language (LCEL) allows you to create <br># arbitrary-length chains with the pipe symbol (|). <br># For instance, if you wanted to format the model’s response, <br># then you could add an output parser to the chain<br>    <br>output_parser = StrOutputParser()<br><br>lipid_chain = prompt_template ||llm | output_parser<br><br>#Here is the context about lipid panel<br>context = &quot;&quot;&quot;<br>Cholesterol, Total<br>Your total cholesterol can be calculated by adding together your HDL, LDL, and 20% of your triglycerides. It is possible to have “normal” total cholesterol and still have very high LDL. This is why it’s important to look at each number to know if you are in the Safe Zone.<br><br>HDL Cholesterol<br>This is sometimes called the “good” cholesterol because higher levels are associated with a lower risk for cardiovascular events. HDL brings excess cholesterol from other areas of the body and returns it to the liver for recycling.<br><br>However, having high HDL is not enough to outweigh the harmful effects of high LDL.<br><br>LDL Cholesterol<br>This number is key. This “bad” cholesterol is what can build up in the arteries and causes heart attacks and strokes.<br><br>LDL is a lipoprotein that carries cholesterol in the blood. It is not actually cholesterol, but it carries cholesterol through the body. When LDL’s job is done, it returns to the liver for recycling. If that recycling process is impaired, LDL can build up in our artery walls.<br><br>Higher LDL levels are known to cause cardiovascular disease. If your LDL is very high, it is likely due to or at least influenced by genetics. Changing your diet can lower LDL, on average by 10-15%. If you have very high LDL, you may need one or more medications to lower it to the Safe Zone. Don’t wait to get there – you can always start medication and work on your diet at the same time.<br><br>The Safe Zone for your LDL varies based on your medical and family history. The American College of Cardiology and American Heart Association 2018 Cholesterol Guidelines recommends lowering LDL to below:<br><br>100 mg/dL for those without existing heart disease.<br>70 mg/dL for someone with heart disease.<br>In 2022 the American College of Cardiology recommended the ideal LDL level for someone with heart disease and additional risk factors is less than 55 mg/dL.<br><br>If your LDL is over 190 mg/dL and you have a family history of early heart attack, heart disease, and/or stroke, you may have an inherited disorder known as familial hypercholesterolemia (FH).<br>&quot;&quot;&quot;<br><br>question = &quot;My LDL is 171, what does it mean?&quot;<br><br>review_chain.invoke({&quot;context&quot;: context, &quot;question&quot;: question})</pre><p>The output of this query would give you a nice, formatted output. You can change the context, system prompt, and human prompt to customize according to your needs and use other prompt templates. You can create a front-end using Streamlit to get interactivity with the user. That’s how you will your own chatbot that you created with LangChain.</p><p>In the next guide, I will cover building a RAG pipeline with LangChain.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1441215e5673" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Make the Most of Asana with These AI features]]></title>
            <link>https://srtpan.medium.com/make-the-most-of-asana-with-these-ai-features-inside-asana-7fccdde0b857?source=rss-47a91e766c89------2</link>
            <guid isPermaLink="false">https://medium.com/p/7fccdde0b857</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[project-management]]></category>
            <category><![CDATA[asana-project-management]]></category>
            <category><![CDATA[asana]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Shruti Pandey]]></dc:creator>
            <pubDate>Mon, 13 Jan 2025 05:59:00 GMT</pubDate>
            <atom:updated>2025-01-13T19:46:05.560Z</atom:updated>
            <content:encoded><![CDATA[<p>Asana is one of my favorite Project Management tools. It has an intuitive interface that enables strong team collaboration with robust task assignment and visibility. Asana makes task management effective with its prioritization features and integration with other applications. You can set up projects to brainstorm, manage tasks, plan resources, strategize operations, etc.</p><p>Asana has new AI features, which make it even more awesome. In this article, I will review three helpful AI features inside Asana that will help you with task management, creating rules, and providing status updates. These features will significantly streamline your workflow, allowing you to focus on finishing your tasks.</p><p>We will focus on<a href="https://help.asana.com/s/article/smart-summaries?language=en_US"> AI-powered task summaries</a>, which provide quick overviews of complex tasks,<a href="https://help.asana.com/s/article/smart-summaries?language=en_US"> AI-generated subtasks</a>, and<a href="https://help.asana.com/s/article/smart-status?language=en_US#:~:text=To%20create%20a%20status%20update%20with%20Asana%20AI%20for%20goals,guidance%20to%20add%20further%20instructions."> Smart Status</a>. These allow you to draft status updates to help identify blind spots, open questions, and roadblocks, and carve an efficient path towards your goals.</p><p>Let’s get started!</p><h3>Create projects in Asana with AI</h3><p>Before we review other features on Asana, let’s get started with setting up the project. You can define the project in the section ‘<em>How do you want to use the project’</em> and be as detailed as possible. This will create tasks inside the project suited for the project that you can edit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*CFdnyoyISUFE1c6F" /><figcaption>A preview of Asana AI feature: Screenshot captured by the author</figcaption></figure><h3>Craft Comprehensive Task Summaries Using AI</h3><p>Asana’s AI can<a href="https://help.asana.com/s/article/smart-summaries?language=en_US"> generate concise task summaries</a> with good precision. Once you have added a project, navigate to the task, add a description, and use AI to summarize the task, draft subtasks, and ask AI to recommend edits to your description.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HCVbY4rwZJhrL2AQ" /><figcaption>A preview of Asana AI feature: Screenshot captured by the author</figcaption></figure><p>When you open a task, if you click the “Summarize” button, the AI analyzes all the information within the task. It considers comments, attachments, and even linked tasks to create a comprehensive overview. You can also ask AI to summarize it in the comments which is so helpful when you have chains of comments in the discussion. A quick summary will help get you back on track and keep the team centered on the task.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KOCEB0C02Z7bTv5VKGKpgg.png" /><figcaption>A preview of Asana AI feature: Screenshot captured by the author</figcaption></figure><p>This feature has saved me quite some minutes that I used to spend manually reviewing and summarizing task details, especially for complex tasks or those with extensive discussions. I am sure it will help you immensely.</p><h3>Add Subtasks feat. AI</h3><p>More than tasks, subtasks consume a lot of time. With the new <a href="https://forum.asana.com/t/create-subtasks-automatically-with-asana-intelligence/547899">Asana Intelligence</a>, be ready to be amazed. The AI creates subtasks based on the main task’s description and context, as you can see here in the screenshot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*dJk4_wz3NTOIF0QX" /><figcaption>A preview of Asana AI feature: Screenshot captured by the author</figcaption></figure><p>I find this especially helpful for breaking down complex projects or when I’m unsure where to start. The AI often suggests subtasks I might have overlooked (hi, fellow multitasker!), ensuring a more thorough approach to task completion.</p><h3>Automate Workflows with Custom AI Rules</h3><p><a href="https://forum.asana.com/t/smart-rules-automate-tasks-and-manage-complex-workflows-at-scale-using-ai/875874">Smart Rules in Asana</a> simplify the workflows by effective resource allocations, task prioritization, and scheduling. You can configure rules that analyze task descriptions and deadlines to adjust priority levels and resolve conflict. I have added an example below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*6meclW_MmLaewKaM" /><figcaption>A preview of Asana AI feature: Screenshot captured by the author</figcaption></figure><p>AI rules can also trigger notifications and updates. In my projects, I sometimes use rules to send alerts when high-priority tasks are nearing their due dates or when dependencies are completed. I also want to mention that it’s crucial to test out these rules in a controlled environment before putting them in production to make sure the project is still streamlined after applying all the rules.</p><h3>Status Update Overwhelm? Use AI!</h3><p>We all sometimes struggle with the overwhelm that comes with providing status updates. It’s a time-intensive process that takes a log of multiple projects we might be working on, and audits all the complex tasks we have finished. You can simplify with the use of AI inside Asana.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*4HGGllUHdWw_6HdE" /><figcaption>A preview of Asana AI feature: Screenshot captured by the author</figcaption></figure><p>Select a date range in which you want to provide an update and be precise with your prompt of what you want in the status report. When I want to go specific, I fine-tune the prompts like <em>“Discuss recent status updates for projects and indicate whether they’re on track, at risk, or off track.”</em></p><p>You can combine the smart status update with smart rules to schedule your status updates to be sent to your project lead or manager at regular intervals. You can use it to send it before weekly team meetings or annual executive reports.</p><p>Along with these, more AI features are being added to Asana with each new update and I cannot wait to dive into them in my next blog posts. Subscribe for more!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7fccdde0b857" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>