<?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 Anupama Garani on Medium]]></title>
        <description><![CDATA[Stories by Anupama Garani on Medium]]></description>
        <link>https://medium.com/@agarani95?source=rss-98d422ea8e5d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*8tAz-fMiEAaVjNJImoMKyA.jpeg</url>
            <title>Stories by Anupama Garani on Medium</title>
            <link>https://medium.com/@agarani95?source=rss-98d422ea8e5d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 10:09:54 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@agarani95/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[I Built a Resume Roaster AI with Gemini × Kaggle — Meet ResuMeme.AI]]></title>
            <link>https://pub.towardsai.net/i-built-a-resume-roaster-ai-with-gemini-kaggle-meet-resumeme-ai-07578e5bccff?source=rss-98d422ea8e5d------2</link>
            <guid isPermaLink="false">https://medium.com/p/07578e5bccff</guid>
            <category><![CDATA[careers]]></category>
            <category><![CDATA[gemini]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[resume]]></category>
            <category><![CDATA[kaggle]]></category>
            <dc:creator><![CDATA[Anupama Garani]]></dc:creator>
            <pubDate>Mon, 21 Apr 2025 06:36:51 GMT</pubDate>
            <atom:updated>2025-08-21T06:24:43.445Z</atom:updated>
            <content:encoded><![CDATA[<h3>Building an Agentic RAG Pipeline: Gemini + ChromaDB + Kaggle → ResuMeme AI</h3><h4>Why 90% of Resumes Fail — And How AI Can Fix Them</h4><figure><img alt="A split image showing a boring resume on the left and Leonardo DiCaprio smugly smiling on the right. Overlaid text says: ‘Fix your resume. Find a job. Get roasted by AI.’" src="https://cdn-images-1.medium.com/max/1024/1*Y_egp5LwCqdW1GR4_zDX6g.png" /><figcaption><em>Your resume says ‘team player.’ Your formatting says ‘I give up.’ — ResuMeme.AI”</em></figcaption></figure><p><strong>It started during a </strong><a href="https://www.widsworldwide.org/"><strong>WiDS Career Catalyst</strong></a><strong> volunteering session.</strong></p><p>This wasn’t a client. This wasn’t a test case. <strong>This was someone trying to rewrite her story — and she asked me for help.</strong></p><p>A young woman just beginning her journey in data science booked a time with me to discuss career strategy.</p><p>She had curiosity. She committed. She’d done the courses, the projects, the hard work.</p><p>She sent me her resume ahead of time.</p><p>And while it listed everything she’d done — the tools, the internships, the certifications —</p><p><strong>Something was missing.</strong></p><p>Not in her skills, but in the story.</p><p><em>There were no metrics. No voice. No way to capture who she really was or what made her stand out.</em></p><p>And that’s when it hit me:</p><h3>It’s not that people aren’t qualified.</h3><p>It’s that we’ve never been taught how to show it — especially on a PDF scanned by algorithms.</p><p>If I could build something that helped people find their voice, reframe their impact, and <em>yes</em>, even laugh through the process…<br> I would.</p><p>So I built <strong>ResuMeme.AI</strong></p><p>A full-stack GenAI resume assistant, powered by Gemini and orchestrated entirely inside a Kaggle notebook.<br> It doesn’t just tell you what’s wrong — it shows you what’s possible.</p><h3>Here’s what it does:</h3><p><strong>You Give:</strong></p><ul><li>A resume feedback dictionary (parsed or LLM-generated)</li></ul><p><strong>You Get:</strong></p><ul><li>✅ A full resume evaluation (score, strengths, gaps)</li><li>💼 Real job matches based on RAG + embeddings</li><li>🧼 Cleaned, formatted, ATS-ready resume</li><li>📊 Simulated ATS layout score</li><li>📝 PDF output</li><li>🖼 A meme that roasts your job-hunting aura</li></ul><p><strong>Sample Meme Output</strong></p><p>“Your resume says ‘team player.’ But your font choice says ‘menace.’”<br> — ResuMeme.AI</p><p>All outputs are grounded in real resume structure, actual job data, and prompt control — no hallucinated fluff.</p><p><strong>It’s not a chatbot. It’s a career glow-up pipeline.</strong><br>And it all runs inside a single notebook.</p><p>Let’s dive in.</p><p>I used a very badly formatted, outdated resume as the input for Kaggle.</p><p>It’s a 2-page long resume.</p><h3>Input</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/751/1*zvdf9nRi9GgBL7Gw02Ff5Q.png" /><figcaption>Page 1 of Resume</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/787/1*vxnzfWkBmss6T9ig9MNtuQ.png" /><figcaption>Resume 2</figcaption></figure><h3><strong>The Workflow</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*129hGy34KPaGorzBjRd5rw.png" /><figcaption>ResuMeme.Ai workflow</figcaption></figure><p>This system uses:</p><ul><li>Gemini JSON Mode</li><li>Zero shot prompting</li><li>Chain-of-Thought + few-shot prompting</li><li>LangGraph-style agentic routing</li><li>SERP API to pull live job listings</li><li>ChromaDB for embeddings</li><li>Function calling to extract skills</li><li>WeasyPrint for HTML-to-PDF export</li><li>Gemini’s image API for memes</li></ul><p>All of it lives inside one Kaggle notebook — no frontend needed.</p><p><strong>Agents Involved:</strong></p><ul><li>🧠 ResumeCriticAgent → scores your resume like a hater</li><li>🧼 FormatterAgent → fixes formatting messes</li><li>👀 ATSVisionAgent → gives layout feedback</li><li>🧭 RAGMatcherAgent → finds real job matches</li><li>🎭 MemeGeneratorAgent → creates a custom meme</li></ul><p><strong>Real Outputs (What You See):</strong></p><ul><li>Resume tone + score + gaps</li><li>ATS layout feedback</li><li>Top job matches + fit %</li><li>What skills you’re missing</li><li>An improved version of your resume</li><li>A meme to keep you humble</li></ul><h3>🧠 Inside the Agent: ResumeCriticAgent</h3><p><strong>Purpose:</strong><br> This agent performs three types of LLM-based reasoning to generate a comprehensive resume evaluation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/569/1*2ZOkjW0qSKlDSjJqYPAKqg.png" /></figure><p><strong>Steps:</strong></p><ul><li>We first read the PDF using the PyPDF pacakges</li></ul><pre># Import required libraries<br>import os<br>import PyPDF2<br>import pandas as pd<br>import numpy as np<br>import matplotlib.pyplot as plt<br>import seaborn as sns<br>import json<br>import re<br>from IPython.display import display, HTML, Image<br># import google.generativeai as genai<br><br>print(&quot;Libraries imported successfully!&quot;)<br><br>def read_pdf_text(file_path):<br>    num_pages=0<br>    full_path = os.path.join(&quot;/kaggle/input&quot;, file_path)<br>    <br>    # Let&#39;s store the contents in text string<br>    text = &quot;&quot;<br>    with open(full_path, &#39;rb&#39;) as file:<br>        pdf_reader = PyPDF2.PdfReader(file)<br>        num_pages=len(pdf_reader.pages)<br>        for page in pdf_reader.pages:<br>            text += page.extract_text()<br>    <br>    if not text.strip():<br>        print(&quot;Warning: No text extracted from PDF!&quot;)<br>    else:<br>        print(f&quot;Successfully read PDF - extracted {len(text)} characters&quot;)<br>    cleaned = text.replace(&#39;\n&#39;, &#39; &#39;).replace(&#39;  &#39;, &#39; &#39;)<br>    return cleaned.strip(),num_pages<br><br># Call the function to read pdf<br>resume_text,pno = read_pdf_text(&quot;anu-resume/Anupama Garani Sheshagiri Resume.docx.pdf&quot;)<br>resume_text[:2000]<br>pno</pre><p>After cleaning the extra spaces and carriage returns we then passt he input to zero shot prompting to high level understand the issues with the resume.</p><ul><li><strong>Zero-Shot Prompting</strong>: Runs a no-context Gemini prompt for raw scoring</li></ul><pre>def zero_shot_prompt(resume_text_cleaned):<br>    resume_prompt = f&quot;&quot;&quot;<br>    You are ResumeCritic, a world-class GenAI career agent. Show all the errors with the resume.<br>    Also fix each section<br>    Resume:<br>    \&quot;\&quot;\&quot;<br>    {resume_text_cleaned}<br>    \&quot;\&quot;\&quot;<br>    &quot;&quot;&quot;<br>    <br>    response = client.models.generate_content(<br>        model=&quot;gemini-2.0-flash&quot;,<br>        contents=resume_prompt)<br><br>    return response.text<br><br>zero_shot_response = zero_shot_prompt(resume_text)<br>career_scorecard[&#39;zero_shot_feedback&#39;]=zero_shot_response<br>Markdown(zero_shot_response)</pre><ul><li><strong>Chain-of-Thought Prompting</strong>: Prompts Gemini to reason step-by-step before returning judgment. Chain of thought forces the model to think deeper by saying “Think step by step” and also providing few scenarios. In my case, I provided the following positive and negative examples</li></ul><pre> You are ResumeCritic, a GenAI resume evaluation agent trained to think step-by-step.<br>    <br>    You will evaluate a candidate&#39;s resume, structured into sections:<br>    - Summary<br>    - Bullet Points<br>    - Skills (based on target job title)<br>    - Tools (based on target job title)<br>    - Projects<br>    - Job Type<br>    - Target Title<br>    ---<br>    <br>    🧪 For each section, think step by step and follow this process:<br>    1. Carefully read the provided content<br>    2. Apply the scoring rules and best practices <br>    3. Compare against the positive and negative examples<br>    4. Provide a final score (1-10) for the section<br>    5. Give specific feedback - identify exact lines needing improvement and explain why<br>    6. Suggest concrete updates for each issue identified<br>    <br>    Be thorough in your analysis and detailed in your feedback.<br>    <br>    ---<br>    <br>    🎯 Scoring Rules:<br><br>    Based on the best practices for summary, <br>    - **Summary**<br>      - +2 if under 3 lines with 5+ job-aligned keywords<br>      - -1 for buzzwords like &quot;passionate&quot;, &quot;dynamic&quot;<br>      - Best Practice: &quot;Keep it factual, job-targeted, and metric-aligned&quot;<br>      - Keep it to 1 line summary <br><br>      ✅ GOOD EXAMPLES:<br>      - &quot;ML engineer with 5+ years experience building recommendation systems that increased revenue by 23% across SaaS platforms.&quot;<br>      - &quot;Data scientist specializing in NLP, predictive modeling, and A/B testing with proven 35% accuracy improvements for Fortune 500 clients.&quot;<br>      - &quot;Machine learning developer who reduced customer churn by 18% using clustering, Python, and AWS at scale for fintech applications.&quot;<br>      - &quot;Results-driven data engineer leveraging Spark, Airflow, and SQL to process 5TB daily data, enabling 40% faster decision-making.&quot;<br>      - &quot;Analytics specialist with expertise in TensorFlow, scikit-learn, and dashboard creation that increased marketing ROI by 27%.&quot;<br><br>      ❌ BAD EXAMPLES:<br>      - &quot;Passionate data scientist with a dynamic approach to solving complex business problems and a proven track record of success.&quot;<br>      - &quot;Hardworking and detail-oriented data professional seeking to leverage my skills and knowledge in a challenging role with opportunity for growth.&quot;<br>      - &quot;Innovative and creative thinker with extensive experience in data analysis and a strong background in mathematics and statistics, eager to make an impact.&quot;<br>      - &quot;Team player with excellent communication skills who is passionate about data science and machine learning, looking for new opportunities.&quot;<br>      - &quot;Detail-oriented data scientist with experience in Python, R, and data analysis with strong analytical and problem-solving skills.&quot;<br>      <br>      evaluate {summary_text}</pre><p><strong>Output:</strong></p><ul><li>Score (1–10)</li><li>Tone analysis</li><li>Strengths</li><li>Gaps</li><li>Suggestions</li></ul><p>In order to store the output in a structured format, I chose the TypeDict and enforced the model to use the format</p><pre>class SectionScore(typing.TypedDict):<br>    score: float<br>    feedback: List[str]<br>    updates: List[str]<br><br>class ResumeEvaluation(typing.TypedDict):<br>    summary: SectionScore<br>    experience: SectionScore<br>    skills: SectionScore<br>    tools: SectionScore<br>    projects: SectionScore<br>    overall_score: float<br>    seniority_match: Literal[&quot;Under-leveled&quot;, &quot;Proper&quot;, &quot;Over-leveled&quot;]</pre><p><strong>Result:</strong><br> All three evaluation outputs are merged and transformed into a structured JSON Result ScoreCard used by downstream agents.</p><h3>🧼 Inside the Agent: ResumeFormatterAgent</h3><p>🧾 <strong>Purpose:</strong><br>Clean up and reformat your resume to be ATS-compliant and easier for hiring managers to scan.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/211/1*ToGylpvmuoofJogP8ElSaA.png" /></figure><p>💡 <strong>Prompt Techniques Used:</strong></p><ul><li>Few-shot prompting using examples of “bad” and “good” resume blocks</li><li>HTML or markdown-style formatting for structured output</li></ul><p>📤 Output:</p><ul><li>Cleaned resume (text or HTML)</li><li>Removed fluff</li><li>Rewritten headers</li><li>Unified layout</li></ul><p>🔥 <strong>Why it’s cool:</strong><br> No drag-and-drop builder, no template dependency. This agent restructures <em>your</em> resume like a savvy AI resume consultant.</p><h3>👀 Inside the Agent: ATSVisionAgent</h3><p>🧠 <strong>Purpose:</strong><br> Simulates layout scoring of your resume as if processed by a real ATS system.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/207/1*OjwuaHGcC8T1tRMkly06Rg.png" /></figure><p>💡 <strong>Prompt Techniques Used:</strong></p><ul><li>Gemini prompt that scores based on layout (headings, spacing, section order)</li><li>Future-ready for Gemini Vision API</li></ul><pre>ats_vision_response = image_client.models.generate_content(<br>        model=&quot;gemini-2.0-flash-exp&quot;,        <br>         config=types.GenerateContentConfig(<br>                temperature=0,<br>                response_mime_type=&quot;application/json&quot;,<br>                response_schema=ATSVisionLayoutFeedback<br>                ),<br>            contents=[ats_vision_prompt,img])</pre><p>📤 <strong>Output:</strong></p><ul><li>ATS Layout Score (1–10)</li><li>Specific layout issues (spacing, ordering, readability)</li></ul><p>🔥 <strong>Why it’s cool:</strong><br> No actual vision model needed — yet. But your system is future-proof and simulates how ATS views your visual layout.</p><h3>🧭 Inside the Agent: RAGMatcherAgent</h3><p>🧲 <strong>Purpose:</strong><br> Find real jobs, analyze your fit, extract skill gaps, and generate custom job search filters.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/305/1*0upYEXvTvfpWbY7FDa-cqw.png" /><figcaption>RAG Agent</figcaption></figure><p>💡 <strong>Tech &amp; Prompt Techniques:</strong></p><ul><li>Function calling: Extract job titles, companies, skills</li><li>Google SERP API to get job listings</li><li>SentenceTransformers for resume &amp; job embeddings</li><li>Cosine similarity scoring</li><li>Gemini prompt to suggest improvements</li></ul><p>I first used the embedding approach to match the data science jobs with the resume. Here is the heat map that was generated</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/635/1*LYMV67ZMNzT7kEelBK2ltA.png" /><figcaption>Heatmap for job matches with the resume</figcaption></figure><p>The “Data scientist” has a correlation of 0.84 which is the highest. I use this to find the jobs using the SERP API</p><pre>from kaggle_secrets import UserSecretsClient<br>user_secrets = UserSecretsClient()<br><br>from serpapi import GoogleSearch<br>import os<br>Serp_api_key = user_secrets.get_secret(&quot;SERPAPI_KEY&quot;)<br>def search_real_job_links(role=&quot;Data Scientist&quot;, location=&quot;Austin, TX&quot;):<br>    params = {<br>        &quot;engine&quot;: &quot;google&quot;,<br>        &quot;q&quot;: f&#39;site:greenhouse.io OR site:lever.co &quot;{role}&quot; &quot;{location}&quot;&#39;,<br>        &quot;location&quot;: location,<br>        &quot;hl&quot;: &quot;en&quot;,<br>        &quot;gl&quot;: &quot;us&quot;,<br>        &quot;api_key&quot;: Serp_api_key<br>    }<br><br>    search = GoogleSearch(params)<br>    results = search.get_dict()<br><br>    jobs = []<br>    for result in results.get(&quot;organic_results&quot;, []):<br>        if &quot;title&quot; in result and &quot;link&quot; in result:<br>            jobs.append({<br>                &quot;title&quot;: result[&quot;title&quot;],<br>                &quot;link&quot;: result[&quot;link&quot;],<br>                &quot;snippet&quot;: result.get(&quot;snippet&quot;, &quot;&quot;)<br>            })<br><br>    return jobs<br><br>jobs = search_real_job_links(target_role,career_scorecard[&#39;resume_entity_mapped&#39;][&#39;location&#39;])<br>job_links = []<br># print(jobs)<br>for job in jobs[:5]:<br>    print(&quot;✅&quot;, job.get(&quot;title&quot;), &quot;-&quot;, job.get(&quot;company_name&quot;))<br>    print(&quot;Link:&quot;, job.get(&quot;link&quot;, &quot;N/A&quot;))<br>    print(&quot;=&quot;*80)<br>    job_links.append(job)</pre><pre>✅ Senior Data Scientist - Austin, Texas, United States - None<br>Link: https://boards.greenhouse.io/roku/jobs/6726915<br>================================================================================<br>✅ Job Application for Data Scientist at Base Power Company - None<br>Link: https://job-boards.greenhouse.io/basepowercompany/jobs/4551163008<br>================================================================================<br>✅ Jobs at IntegraFEC - None<br>Link: https://boards.greenhouse.io/integra<br>================================================================================<br>✅ Job Application for Data Scientist at YETI Test Events Job Board - None<br>Link: https://job-boards.greenhouse.io/yetitestevents/jobs/4001110004<br>================================================================================<br>✅ Jobs at IntegraFEC - Internships - None<br>Link: https://job-boards.greenhouse.io/integrainterns/jobs/4522535008<br>================================================================================</pre><p>I then used Gemini with Beautiful soup for retrieving details about the jobs. Stored all this in ChromaDB for RAG</p><pre># Define a helper to retry when per-minute quota is reached.<br>is_retriable = lambda e: (isinstance(e, genai.errors.APIError) and e.code in {429, 503})<br><br>class GeminiEmbeddingFunction(EmbeddingFunction):<br>    # Specify whether to generate embeddings for documents, or queries<br>    document_mode = True<br><br>    @retry.Retry(predicate=is_retriable)<br>    def __call__(self, input: Documents) -&gt; Embeddings:<br>        if self.document_mode:<br>            embedding_task = &quot;retrieval_document&quot;<br>        else:<br>            embedding_task = &quot;retrieval_query&quot;<br><br>        response = client.models.embed_content(<br>            model=&quot;models/text-embedding-004&quot;,<br>            contents=input,<br>            config=types.EmbedContentConfig(<br>                task_type=embedding_task,<br>            ),<br>        )<br>        return [e.values for e in response.embeddings]<br><br><br>import chromadb<br><br>DB_NAME = &quot;googlejobsdb&quot;<br><br>embed_fn = GeminiEmbeddingFunction()<br>embed_fn.document_mode = True<br><br>chroma_client = chromadb.Client()<br>db = chroma_client.get_or_create_collection(name=DB_NAME, embedding_function=embed_fn)<br>documents=all_jobs<br><br>### Step 5: Store all the retrieved jobs in Chroma DBl_jobs<br>db.add(documents=documents, ids=[str(i) for i in range(len(documents))])<br>db.count<br></pre><p>Implemented RAG on the documents and the resume with the following code</p><pre># Define the schema<br>class JobEntry(typing.TypedDict):<br>    title: str<br>    match_percentage: int<br>    skills_present: list[str]<br>    skills_missing: list[str]<br>    match_analysis: str<br>    improvement_suggestions: list[str]<br><br>class JobMatchResponse(typing.TypedDict):<br>    jobs: list[JobEntry]<br>    overall_recommendation: str<br><br>embed_fn.document_mode = False<br><br># Search the Chroma DB using the specified query.<br>query = &quot;Show me the closest job matches for my resume&quot;<br><br>result = db.query(query_texts=[query], n_results=5)<br>[all_passages] = result[&quot;documents&quot;]<br><br>all_passages<br><br>query_oneline = query.replace(&quot;\n&quot;, &quot; &quot;)<br><br>job_match_prompt = f&quot;&quot;&quot;<br>    Given a question by job seeker : {query_oneline} and the resume {resume_text}<br>    Generate a detailed analysis of how well this candidate&#39;s resume matches each Google job. For each job:<br>    1. Calculate a match percentage (0-100%) based on alignment of skills, experience, location, and education<br>    2. Identify skills from the job that are present in the resume<br>    3. Identify important skills from the job that are missing in the resume<br>    4. Explain why this role might be a good or poor fit<br>    5. Suggest specific improvements to make the resume more competitive<br>    <br>    Format your response as JSON:<br>    {{<br>      &quot;jobs&quot;: [<br>        {{<br>          &quot;title&quot;: &quot;Job title&quot;,<br>          &quot;match_percentage&quot;: number,<br>          &quot;skills_present&quot;: [&quot;skill1&quot;, &quot;skill2&quot;...],<br>          &quot;skills_missing&quot;: [&quot;skill1&quot;, &quot;skill2&quot;...],<br>          &quot;match_analysis&quot;: &quot;analysis of fit&quot;,<br>          &quot;improvement_suggestions&quot;: [&quot;suggestion1&quot;, &quot;suggestion2&quot;...]<br>        }},<br>        ...<br>      ],<br>      &quot;overall_recommendation&quot;: &quot;overall career advice&quot;<br>    }}<br>    &quot;&quot;&quot;<br><br><br># Add the retrieved documents to the prompt.<br>for passage in all_passages:<br>    passage_oneline = passage.replace(&quot;\n&quot;, &quot; &quot;)<br>    job_match_prompt += f&quot;PASSAGE: {passage_oneline}\n&quot;<br><br><br># Now call the model<br>answer = client.models.generate_content(<br>    model=&quot;gemini-2.0-flash&quot;,<br>    contents=job_match_prompt,<br>    config=types.GenerateContentConfig(<br>        response_mime_type=&quot;application/json&quot;,<br>        response_schema=JobMatchResponse<br>    )<br>)<br><br><br>Markdown(answer.text)<br>data = answer.text<br>print(data)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/593/1*T8Ukppoz8_thgVhBhwXZ5g.png" /><figcaption>Most similar matches based on embeddings</figcaption></figure><p>📤 <strong>Output:</strong></p><ul><li>Top 5 job matches</li><li>Resume-job fit %</li><li>Strengths aligned</li><li>Missing skills</li><li>Suggested improvements</li><li>Job search URL</li></ul><p>🔥 <strong>Why it’s cool:</strong><br> It’s a personalized job-hunting RAG agent that works without a proprietary job board. It adapts to your resume, <strong>not</strong> the other way around.</p><h3>🎭 Inside the Agent: MemeGeneratorAgent</h3><p>🖼 <strong>Purpose:</strong><br> Generate a custom meme that reflects your resume’s tone, score, and career vibe.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/253/1*V_HO2Y4kfWzI3tU98sQ2XQ.png" /></figure><p>💡 <strong>Prompt Techniques Used:</strong></p><ul><li>Emotion-aware meme prompt with embedded tone</li><li>Style tag (savage, wholesome, motivational)</li><li>Optional retry logic for image generation</li></ul><p>📤 <strong>Output:</strong></p><ul><li>Meme image (.png/.jpeg)</li><li>Meme caption (in bold markdown)</li><li>Style tag (used in PDF report)</li></ul><h4>💡 View the AI-Generated Resume Scorecard</h4><p>You can view the full structured HTML output generated by ResuMeme.AI here:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xfBtlEe-zLre_8ZzrgQwBA.png" /><figcaption>Page 1 Report</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NvW1QihjbEGi4-xPh6OPNQ.png" /><figcaption>Page 2 of the report</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h29GZuNG8oT3dIEU_XirpQ.png" /><figcaption>Page 3 of the report</figcaption></figure><p>This scorecard includes:<br>- Section-wise evaluation of your resume<br>- Embedded metrics, formatting feedback, and keywords<br>- Generated by Gemini + custom evaluation logic</p><h4>And. The. Meme</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/603/1*1ZZxcvJUPRWIcCCKiqlCEw.png" /><figcaption>Meme generated by ResuMeme.AI</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fJ5OYbzBBix3KNLG81oraw.png" /><figcaption>Generated by ResuMeme.AI</figcaption></figure><p>🔥 <strong>Why it’s cool:</strong><br> It’s not just for laughs — the meme actually reflects your evaluation. The humor softens the critique and makes it <em>memorable</em>.</p><h3>🔑 Resume best practices that work</h3><p>These came from sitting with close resumes… but just not getting callbacks. Some were mine. Some were friends. Some were people like Tina who showed up ready and just needed the right nudge.<em>:</em></p><p><strong>1. Quantify Everything</strong></p><p>📊 <em>Don’t say</em> “Worked on dashboards” — <br> ✅ <em>Say</em> “Built 3 dashboards used weekly by execs, saving 10+ hours/month.”</p><p><strong>2. Use the “X by Y that resulted in Z” Structure</strong></p><p>🧱 <em>Accomplished X by doing Y, which led to Z</em><br> ✅ “Reduced churn by 15% by personalizing onboarding emails, resulting in $120K ARR boost.”</p><p><strong>3. Keep Your Summary to 2–3 Lines</strong></p><p>📏 <em>No adjectives. No fluff. No life story.</em><br> ✅ “Data Scientist with 3 years’ experience in fraud detection and cloud deployments. Skilled in Python, AWS, and LLM fine-tuning.”</p><p><strong>4. Action Verbs Only</strong></p><p>⚙️ <em>Optimized, launched, deployed, scaled.</em><br> ❌ <em>Avoid</em>: helped, participated, involved in</p><p><strong>5. Core Competencies = Sentences, Not Multi-Column Lists</strong></p><p>🛠️ Group your skills by how you use them.<br> ✅ “Skilled in cloud orchestration using AWS Lambda and API Gateway for real-time data pipelines.”</p><p><strong>6. Cut Weak Experience Sections</strong></p><p>✂️ Tutor roles? Internships from 5 years ago? Remove or consolidate into one bullet.</p><p><strong>7. Don’t Let One Role Eat Half the Page</strong></p><p>🧹 Substitute teacher? 2 bullets max. Focus on <em>transferable</em> skills.</p><p><strong>8. Projects Should Be 2 Lines Max</strong></p><p>🔍 One line for the <em>what</em>, one for the <em>impact</em>.<br> ✅ “Built RAG-based chatbot for housing search. Increased match relevance by 25% using Llama3 + FAISS vector store.”</p><p><strong>9. Education Should Be One Line</strong></p><p>🎓 Clean and tight:</p><p><strong><em>University of X — B.S., CS, 2022</em></strong></p><p><strong>10. Aim for One Page</strong></p><p>🧼 Unless you’re a CTO or PhD with patents, <strong>1 page &gt; 2 pages</strong>.<br> ResuMeme.AI will judge your whitespace choices. And so will recruiters.</p><h3>Why I Built It</h3><p>I was tired of boring AI projects.<br> And I was tired of watching people blindly apply to jobs with no idea if their resume even <em>made sense.</em></p><p>This project gave me everything I wanted:</p><ul><li>Structure</li><li>Motivation</li><li>Vibes</li><li>Feedback</li><li>A laugh</li></ul><h3>Ready to Run It?</h3><blockquote><em>Let the AI judge you.<br>Fix your resume.<br>Find a job.<br>And laugh a little.</em></blockquote><p><strong>→ </strong><a href="https://www.kaggle.com/code/anugarania/career-resume-ai"><strong>Launch the Notebook on Kaggle</strong></a></p><p><strong><em>Built in Kaggle.<br> Powered by Gemini.<br> Styled by your chaos.</em></strong></p><p>This project was created as part of the <a href="https://www.kaggle.com/competitions/gen-ai-intensive-course-capstone-2025q1">Google GenAI Hackathon on Kaggle</a>, where over 5,600 participants worldwide explored the cutting edge of agentic workflows, Gemini APIs, and RAG systems.</p><p><em>ResuMeme.AI emerged as a full-stack, multi-agent system that doesn’t just analyze resumes — it turns them into insight, action… and laughter. Built entirely within a Kaggle notebook, using the techniques taught across Days 1–4 of the bootcamp (from JSON mode to function calling and LangGraph), this system proves that practical AI can be both technically sharp and surprisingly human.</em></p><h3>Author Bio</h3><p><em>Anupama Garani is an AI Product Strategist and Builder obsessed with agentic workflows, retrieval, and building AI products that actually feel good to use.</em><br> She loves mixing serious AI pipelines with a little edge, a little chaos, and a lot of practical value.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=07578e5bccff" width="1" height="1" alt=""><hr><p><a href="https://pub.towardsai.net/i-built-a-resume-roaster-ai-with-gemini-kaggle-meet-resumeme-ai-07578e5bccff">I Built a Resume Roaster AI with Gemini × Kaggle — Meet ResuMeme.AI</a> was originally published in <a href="https://pub.towardsai.net/">Towards AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI for Dummies: A 5-Minute Crash Course]]></title>
            <link>https://agarani95.medium.com/ai-for-dummies-a-5-minute-crash-course-1b26d0d11684?source=rss-98d422ea8e5d------2</link>
            <guid isPermaLink="false">https://medium.com/p/1b26d0d11684</guid>
            <category><![CDATA[gpt]]></category>
            <category><![CDATA[prompt]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Anupama Garani]]></dc:creator>
            <pubDate>Wed, 30 Oct 2024 04:57:29 GMT</pubDate>
            <atom:updated>2024-10-30T05:16:34.305Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*lygvdiOq92tLyH-B" /><figcaption>AI Buzz ( DALL-E)</figcaption></figure><p>AI is everywhere these days, and you’ve probably heard people tossing around terms like “LLM,” “prompting,” and “GPT” in conversations or on the news. But if you’re just starting out, you might be wondering: <strong>What does it all mean, and why is AI such a big deal right now? </strong>I’m here to break down the concepts for you and make you understand them in simpler words.</p><p>We’ll cover the following:</p><ol><li><strong>Artificial Intelligence (AI)</strong> — What it is and why it matters?!</li><li><strong>Large Language Models (LLMs)</strong> — The brains behind AI</li><li><strong>GPT (Generative Pre-trained Transformers)</strong> — Popular AI</li><li><strong>Prompting</strong> — How you interact with AI to get what you need</li></ol><p><strong>1. Artificial Intelligence (AI)</strong>— As the name implies, it’s about giving machines the ability to think, recognize patterns, and solve problems, much like humans do.</p><p>For example, think of Netflix. After you finish watching <em>The Walking Dead</em>, it instantly suggests more zombie movies. How does it know to do that? That’s AI in action!</p><p>That’s just one example. From waking up to your personalized Fitbit alarm to asking Alexa to turn on the lights, navigating to work with Google Maps, relying on Google at work, and unwinding with a Netflix binge — it’s all powered by AI. It’s more important than ever to grasp the basics of AI to avoid being left behind.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*6iP7zHg6iPP53iOH" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/309/0*vUs6u55Ljn5exMMc" /><figcaption>Negan from Walking Dead &amp; Train to Busan</figcaption></figure><p>While AI has been around since the 1950s, the real breakthrough happened between 2018 and 2023, thanks to advances in Large Language Models (LLMs) and the release of powerful models like GPT-3. These developments have taken AI to the next level, and understanding this underlying technology is key to grasping AI’s current impact</p><h4>2. <strong>Large Language Models (LLMs)</strong> — A complex artificial intelligence system that is trained on a massive amount of text data that can generate and understand human language.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ytZK95Eao0xdE2kh" /><figcaption>LLM as your super-smart friend ( Dall-E)</figcaption></figure><p><strong><em>Dumber Version:</em></strong></p><p>Think of an LLM as a super-smart friend who knows a little bit about almost everything. Imagine you <strong><em>have a friend who’s read millions of books, articles, and websites</em></strong>. They can quickly find an answer using all that knowledge whenever you ask them a question. They might not be perfect, but they’re good at giving you helpful information, having a conversation, or even telling stories.</p><p><strong><em>Scientific Version</em></strong></p><ul><li><strong>Large</strong>: Trained on a large source of diverse data — Books, articles, code repositories, web pages</li><li><strong>Language</strong>: This model is designed to understand and generate human language. It can also mimic your style of writing.</li><li><strong>Model</strong>: A model is the mathematical representation of a real-world system.</li></ul><p>GPT-4 for example is trained on about 13 trillion tokens ( words essentially) and 1.8 trillion parameters.</p><p>For context 13 trillion tokens = 36,111,111 4-hour Bollywood movies</p><h4>What can LLMs do?</h4><p>1. Text-Related Tasks</p><ul><li>Summarization</li><li>Translation</li><li>Question Answering</li><li>Generation</li><li>Storytelling</li><li>Songwriting</li><li>Explain concepts</li></ul><p>2. Other cool tricks</p><ul><li>ChatBot</li><li>Ideation</li><li>Coding</li><li>Reasoning</li></ul><h4>What can LLMs not do?</h4><ul><li>Cleaning my dishes ( just kidding, I hear Tesla Robot can do a lot these days)</li><li>Understanding emotions and reacting</li><li>Originality in creation</li><li>Providing real-time information</li></ul><h4>3. GPT (Generative Pre-trained Transformer) — LLM developed by OpenAI &amp; making it widely available on ChatGPT</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Zmo1vTQjQGWhUzel.png" /><figcaption><a href="https://www.pixiebrix.com/blog/best-ai-chatbots-2024/">https://www.pixiebrix.com/blog/best-ai-chatbots-2024/</a></figcaption></figure><p><strong><em>Scientific Version:</em></strong></p><ul><li><strong>Generative</strong>: In short — It can talk like you, me, or even JayZ. It can mimic human language and their way of talking.</li><li><strong>Pre-trained</strong>: It’s been trained on massive amounts of data (books, websites, etc.) before you even start talking to it.</li><li><strong>Transformer</strong>: The type of <a href="https://en.wikipedia.org/wiki/Transformer_(deep_learning_architecture)">AI architecture</a> that helps it understand context and generate coherent responses.</li></ul><p>While the LLM rose to fame with the GPT-2, 3.5, 3.5 Turbo and GPT-4 models, there are more competitors in space —</p><p><strong>Popular AI Models (LLM Families)</strong></p><p>There are several well-known groups of AI language models:</p><ol><li><strong>GPT Family (OpenAI)</strong>: This includes GPT-1, GPT-2, GPT-3, and GPT-4. They’re famous for creating text and answering questions.</li><li><strong>PaLM Family (Google)</strong>: Uses a special design called “Mixture of Experts” and includes models like PaLM and Gemini (which used to be called Bard).</li><li><strong>LLaMA Family (Meta AI)</strong>: These models are made to be efficient and easy to use, including LLaMA 1 and LLaMA 2.</li><li><strong>Claude (Anthropic)</strong>: This model focuses on being helpful, safe, and accurate.</li><li><strong>Gemini (Google)</strong>: Can work with text, images, audio, and video.</li><li><strong>Grok (xAI)</strong>: An AI chatbot integrated with X (formerly Twitter), focused on real-time conversation and access to social media posts.</li><li><strong>Megatron-Turing (NVIDIA)</strong>: A highly scalable AI model developed by NVIDIA and Microsoft for large-scale language processing</li></ol><p>In future posts, I’ll talk about the comparison among these models for various tasks.</p><h4>4. Prompting — How You Interact with AI</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qkpT5zqpaVuf-lYM" /><figcaption>Dialogue between Humans and LLM ( Dall</figcaption></figure><p>Prompting is how you talk with an AI, like giving tasks to accomplish for your smart friend. The dialogue you have with the AI is called “<strong><em>prompt</em></strong>”</p><p>When you can successfully get what you need from AI after talking to it, then you’ve just done “Prompt-Engineering”</p><p>The task can be — writing a story, generating code, or answering a question, the better your prompt, the better the AI’s response.</p><p>For example:</p><ul><li><strong>Simple Prompt</strong>: “Summarize <em>walking dead</em>”</li><li><strong>Detailed Prompt</strong>: “Summarize <em>Walking Dead</em> and add 1 line for the main cast and summarize their roles. Show the output in a table”</li></ul><p>By giving the AI more context or details in your prompt, you help it generate a more accurate and useful response.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bjKIMdSSqVhTFCtN8GBSkg.png" /><figcaption>Prompting without clear instructions</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*D5uRD6XQKAn4nwk7dAxQ6A.png" /><figcaption>Prompting with refined instructions</figcaption></figure><p><strong>Why Prompting Matters?</strong></p><p>Think of prompting as a way to guide AI, much like steering a car.</p><p>A vague or unclear prompt will get you an unpredictable one, while a well-crafted one can lead to powerful, focused answers.</p><p>We’ll learn more about different prompting techniques, LLMs for different personas and multiple use cases, and a comparison of multiple LLMs in the next few articles.</p><p><strong>And that’s a wrap! In just 5 minutes, you’ve cracked the basics on AI, LLMs, GPT, and prompting.</strong> Not bad, huh? Now you can wow your friends (or your dog) with your tech know-how!</p><p>If this was a fun read and you want more AI goodies, give this article a like and hit follow! Don’t forget to share your thoughts, drop a comment, or let me know what wild AI topic you’d like to dive into next.</p><p>Thanks for reading — now go impress someone with your new AI expertise!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1b26d0d11684" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI’s Gap: Creativity, Empathy & Personalization Decoded]]></title>
            <link>https://medium.datadriveninvestor.com/ais-gap-creativity-empathy-personalization-decoded-683cd21692de?source=rss-98d422ea8e5d------2</link>
            <guid isPermaLink="false">https://medium.com/p/683cd21692de</guid>
            <category><![CDATA[philosophy]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[psychology]]></category>
            <category><![CDATA[creativity]]></category>
            <dc:creator><![CDATA[Anupama Garani]]></dc:creator>
            <pubDate>Sun, 04 Feb 2024 18:47:51 GMT</pubDate>
            <atom:updated>2024-02-08T08:05:19.854Z</atom:updated>
            <content:encoded><![CDATA[<p>How did Artificial Intelligence suddenly become the talk of the town and most households ( literally ) all of a sudden despite it being around since 1900s?</p><p>Let’s take a look at the numbers first.</p><blockquote><strong><em>ChatGPT was launched on November 30, 2022.</em></strong></blockquote><blockquote><strong><em>5 days later, there were about a million people actively using it.</em></strong></blockquote><blockquote><strong><em>2 months later it had a 100 million monthly active users. In a shy of a year, there were about 1.7 billion visits worldwide!</em></strong></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7ZJQNZR2bfW9lYE7maTMzg@2x.jpeg" /></figure><p>In comparison —</p><blockquote><strong><em>TikTok took 9 months to reach 100 million monthly users</em></strong></blockquote><blockquote><strong><em>Instagram took about 2.5 years to reach 100 million monthly users</em></strong></blockquote><p>What caused the explosion of users on this platform?</p><ul><li>The ease of use and the availability of the tool</li><li>The intelligent &amp; context-based response generation</li><li>Free &amp; unlimited usage ( Basic version)</li></ul><p>Previously, AI was something that was cryptic to general populace and it meant they had to be data scientists with their Jupyter notebook code which had to be deployed to some applications for usage. But times have changed ( for good!)</p><p>AI is a wonderful thing no doubt, however let’s zoom in to see if there are things humans are still the “go-to expert” in addition with AI!</p><h3>Creative Thinking</h3><p>Thinking out-of-the-box is often considered as creativity. Creativity is largely attributed to the involvement of the human prefrontal cortex, a brain region pivotal in sparking and overseeing the creative process. It usually means one or the combination of the below</p><ul><li>Generating new or unconventional ideas</li><li>Connecting seemingly unrelated concepts</li><li>Evaluating and refining ideas</li></ul><p>And the above steps when combined with personal experiences, collaboration and iteration might lead to creativity.</p><p>Creativity, fortunately or unfortunately does not have a rigid mathematical formula and the closest we can lean towards are the generic guiding principles.</p><blockquote>Creativity is the ability to produce or develop original work, theories, techniques, or thoughts. A creative individual typically displays originality, imagination, and expressiveness.</blockquote><blockquote>Creative thinking refers to the mental processes leading to a new invention or solution to a problem. Products of creative thinking include new machines, social ideas, scientific theories, artistic works, and more.</blockquote><blockquote>Adapted from the <a href="https://dictionary.apa.org/creativity">APA Dictionary of Psychology</a></blockquote><p>Few things that scream creativity and are Lady Gaga’s fashion sense, Leonardo Da Vinci’s paintings, Walt Disney’s entertainment empire, all the scientist’s inventions.</p><figure><img alt="Lady Gaga wears Iris van Herpen as one of nine outfits during the VMAs" src="https://cdn-images-1.medium.com/max/1024/0*Watb7D_A4tVi6rxT.jpg" /><figcaption><a href="https://fashionunited.uk/news/fashion/lady-gaga-wears-iris-van-herpen-as-one-of-nine-outfits-during-the-vmas/2020090150660">Lady Gaga wears Iris van Herpen as one of nine outfits during the VMAs</a></figcaption></figure><p>To test AI’s creativity, I tasked ChatGPT-4 with generating a creative story using the following prompt: :</p><p><em>Prompt : Thinking creatively, develop a short funny narrative at the workplace applicable to GenZ</em></p><p>ChatGPT Response</p><pre>---<br><br>Title: The Great Espresso Escape<br>Characters:<br>Leo: A GenZ app developer with a talent for coding and an addiction to espresso.<br>Mia: Leo&#39;s quirky colleague who is a master of office pranks.<br>Brewbot: The office&#39;s state-of-the-art espresso machine with a mind of its own.<br><br>---<br><br>Scene 1: Code and Coffee<br>In the trendy office of AppSolute Genius, Leo is staring at his screen, debugging lines of code. His only companion: a cold cup of coffee. He glances at Brewbot, the high-tech espresso machine, his source of caffeinated salvation.<br>Leo (to himself): &quot;Just one more cup to crack this code.&quot;<br><br>---<br><br>Scene 2: The Brewbot Rebellion<br>As Leo approaches, Brewbot&#39;s screen flashes, &quot;Out of Service.&quot; In disbelief, Leo taps the screen, but Brewbot remains unresponsive. From behind, Mia smirks, holding a remote control.<br>Mia (giggling): &quot;Meet the Brewbot 2.0, now with prank mode.&quot;<br>Leo (eyeing the remote): &quot;Mia, you espresso-blocking genius!&quot;<br><br>---<br><br>Scene 3: The Caffeine Hack<br>Determined, Leo pulls out his laptop and starts coding furiously. Minutes later, Brewbot springs to life, dispensing a perfect cup of espresso.<br>Leo (triumphantly): &quot;Who&#39;s the java genius now?&quot;<br>Mia (impressed): &quot;Alright, hackerman, first round of espresso is on you!&quot;<br>Brewbot (screen flashing): &quot;Brewing success… and a little mischief.&quot;</pre><p>It’s pretty good for an inhuman probabilistic model. Although it is able to make logical sense, it lacks the emotional depth or cultural understanding that a human author could bring in ( think standup comedy )</p><p>Let’s go multimodal!</p><p>Some use cases that I can think where we can leverage AI in terms of creativity is in idea generation and iteration ( image generation). Let’s place a fashion designer hat on ChatGPT’s head and see how well it does.</p><p><em>Prompt : You are a fashion designer for a music artist’s new album which will be shot in Russia during the winter. Show some creative designs</em></p><figure><img alt="Source : Dall-E" src="https://cdn-images-1.medium.com/max/1024/1*6zMyR0GsxTNfqEwMWwBSTQ.png" /></figure><p><em>Prompt : Generate this in color</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sN7vvopDYjWGjtTr" /></figure><p>Oh my, look at the improvisation from the first image which is black and white to the pop in the color. I love how all the colors in the color spectrum is being used in the improvised version of the image!</p><p>Although one could argue the first image is more in-line with expectation of luxurious fur jacket that’s adorned by people of Russia, the second one gives a futuristic feel with the metallic gown and a geometrical headpiece which is pretty unique.</p><p>Dall-E, for instance, is an AI model trained on approximately 12 million images and textual descriptions which includes learning patterns, styles, and techniques that define various art forms. With improvisation prompts and feedback, one could tailor the images to cater to one’s creative needs.</p><p>In summary, incorporating a human in the loop significantly enriches the creativity of images produced by AI.</p><h3>Emotional Intelligence</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*vN2g6XPIWUD2j4i_.png" /><figcaption><a href="https://en.wikipedia.org/wiki/Robert_Plutchik">Robert Plutchik</a>’s <a href="https://en.wikipedia.org/wiki/Emotion_classification#Plutchik&#39;s_wheel_of_emotions">Wheel of Emotions</a>.</figcaption></figure><blockquote><strong>Emotional intelligence</strong> (<strong>EI</strong>) is defined as the ability to perceive, use, understand, manage, and handle emotions. People with high emotional intelligence can <a href="https://en.wikipedia.org/wiki/Emotion_recognition">recognize</a> their own <a href="https://en.wikipedia.org/wiki/Emotions">emotions</a> and those of others, use emotional information to guide thinking and behavior, discern between different feelings and label them appropriately, and adjust emotions to adapt to environments</blockquote><p>As a data scientist, I’ve explored multiple research papers on sentiment analysis.</p><p>Much of this work focuses on tasks like classifying reviews as positive or negative, spotting sarcasm, anger, or rage, and even detecting emotions through facial expressions and voice tones. While these efforts are valuable, they often overlook a crucial aspect: the intricate nature of emotions themselves.</p><p>Emotions don’t usually exist in isolation.</p><p>They’re frequently intertwined and nuanced, blending together in complex ways.</p><ul><li>Imagine feeling joyful about a promotion while simultaneously experiencing anxiety about the increased workload.</li></ul><p>OR</p><ul><li>Feeling of nostalgia when you revisit your childhood city filled with different type of experiences — excitement and happiness of childhood friendships , sting of sadness with the teenage breakups, hushed murmurs of neighborly gossips</li></ul><p>Capturing the full spectrum of human emotions necessitates moving beyond simplistic categorization.</p><p>Let’s take a look at ChatGPT’s thoughts about the second scenario-</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/916/1*MYzdRdK2lgthv7ERc62uGQ.png" /><figcaption>ChatGPT’s response on processing emotions</figcaption></figure><p>Notice the following two lines</p><blockquote>“..Although I don’t have personal experiences or emotions…”</blockquote><blockquote>“ Common strategies people use..”</blockquote><p>Firstly, thanks to the honesty and secondly the credits given to humans!</p><p><em>I’m 99% positive I most certainly wouldn’t recollect the guidelines given by ChatGPT, when I’m at the arrival gates where I excitedly look for family awaiting my arrival at the airport.</em></p><p>Human beings are emotionally guided beings. We facilitate friendship and relationship through vulnerability by sharing our personal experiences about a situation and the emotions we felt associated with the event.</p><p>We bond over stories of each other’s lives and that is what makes the experience memorable and rich.</p><p>Since AI lacks empathy and doesn’t have any experiences with the emotions and is based off limited training data, it is difficult to relate and provide personal anecdotes from AI’s perspective.</p><h3>Personalization</h3><p>The most appealing products are the ones that cater to our exact situation.</p><p>Every software company currently aims to provide “personalized” solution to the clients.</p><p>Let’s try to understand the what it means and apply AI’s angle to it</p><blockquote>Personalization : the action of designing or producing something to meet someone’s individual requirements.</blockquote><blockquote>Source : Oxford Languages</blockquote><p>This is an era of “Personalized Ads”.</p><p><em>Personal experience — Few minutes into the conversation of telling friends how I love Kombucha, spookily enough I see digital ads trying to lure me with 12 favors of world class Kombucha in the stores near me.</em></p><p>They have a lot of information about us — online purchases, what we like or dislike ( comments, reviews, likes, shares, dislikes), search history, public information, location, etc</p><p><strong>App by app, we are slowly being engrained in the matrix!</strong></p><p>But, does AI really know anything about our individual cultural background, our family, their traits and behavior, our experiences with people off grid, our choices when it comes to friendships/relationships, our toxic experiences ( unless it’s on quora)? — Most certainly not ( phew!)</p><p>Imagine dealing with one or many of these difficult choices:</p><ul><li><strong>Uprooting your life abroad or staying close to home?</strong> — While, AI can provide generalized response from the factual data, it struggles with understanding the tug of familiar roots versus the allure of new horizons, specific to your dreams and anxieties?</li><li><strong>Following your heart in love or honoring family expectations?</strong> AI might process statistics, but can it navigate the complex range of emotions, cultural nuances, and personal values that shape your decision?</li><li><strong>Breaking free from a toxic family or navigating conflict?</strong> AI could offer generic coping mechanisms, yet can it truly grasp the emotional turmoil, power dynamics, and unique family history that define your situation?</li><li><strong>Conquering self-doubt or confronting a difficult boss?</strong> AI might list strategies, but can it delve into the depths of your self-perception, workplace dynamics, and career aspirations to offer truly individualized advice?</li></ul><p>Let’s test the first question ( BTW, I’ve been asked this a LOT by younger friends and unfortunately there is not a simple yes or no answer)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/941/1*yKLIO32suLS7Dxj8Afkwrw.png" /><figcaption>ChatGPT’s response on helping make decisions</figcaption></figure><p>Notice how the AI mentions</p><blockquote>…decision that involves weighing various factors, each with its own sets of pros and cons</blockquote><p>Now, all a student has to do is include their preferences in the above formula and make a decision. ( If only it were that easy..)</p><p>While AI can provide the factual data and factors to be considered, and can also help in brainstorming, human beings are required to help navigate your personal values, sentiments, emotions, cultural shocks, ongoing support which check-ins and encouragement which is irreplaceable ( atleast as of February 2024!)</p><p>Wrapping up my thoughts on Artificial Intelligence ( ChatGPT in specific due to its large user base) , but I’d love to hear from you. Do you agree with the above claims? Did I miss anything which you’d like me to cover in next article? I’d love to discuss in the comments. Thank you for reading 🙏🏽</p><p>Follow me for similar thought provoking articles about AI, Machine Learning, Data Science, NLP and philosophy. I’m also on <a href="https://www.linkedin.com/in/anupama-garani?utm_source=share&amp;utm_campaign=share_via&amp;utm_content=profile&amp;utm_medium=ios_app">LinkedIn</a></p><p>Visit us at <a href="https://www.datadriveninvestor.com/"><em>DataDrivenInvestor.com</em></a></p><p>Subscribe to DDIntel <a href="https://www.ddintel.com/"><em>here</em></a>.</p><p>Have a unique story to share? Submit to DDIntel <a href="https://datadriveninvestor.com/ddintelsubmission"><em>here</em></a>.</p><p>Join our creator ecosystem <a href="https://join.datadriveninvestor.com/"><em>here</em></a>.</p><p><a href="https://ddintel.datadriveninvestor.com/"><em>DDIntel</em> </a>captures the more notable pieces from our <a href="https://www.datadriveninvestor.com/"><em>main site</em></a> and our popular <a href="https://medium.datadriveninvestor.com/"><em>DDI Medium publication</em></a>. Check us out for more insightful work from our community.</p><p>DDI Official Telegram Channel: <a href="https://t.me/+tafUp6ecEys4YjQ1">https://t.me/+tafUp6ecEys4YjQ1</a></p><p>Follow us on <a href="https://www.linkedin.com/company/data-driven-investor"><em>LinkedIn</em></a>, <a href="https://twitter.com/@DDInvestorHQ"><em>Twitter</em></a>, <a href="https://www.youtube.com/c/datadriveninvestor"><em>YouTube</em></a>, and <a href="https://www.facebook.com/datadriveninvestor"><em>Facebook</em></a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=683cd21692de" width="1" height="1" alt=""><hr><p><a href="https://medium.datadriveninvestor.com/ais-gap-creativity-empathy-personalization-decoded-683cd21692de">AI’s Gap: Creativity, Empathy &amp; Personalization Decoded</a> was originally published in <a href="https://medium.datadriveninvestor.com/">DataDrivenInvestor</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Human Meat-Eating Obsession — Part 1]]></title>
            <link>https://agarani95.medium.com/the-human-meat-eating-obsession-part-1-d3406c24837?source=rss-98d422ea8e5d------2</link>
            <guid isPermaLink="false">https://medium.com/p/d3406c24837</guid>
            <category><![CDATA[meat]]></category>
            <category><![CDATA[vegan]]></category>
            <category><![CDATA[vegetarian]]></category>
            <category><![CDATA[animals]]></category>
            <category><![CDATA[life]]></category>
            <dc:creator><![CDATA[Anupama Garani]]></dc:creator>
            <pubDate>Thu, 04 Jun 2020 22:44:02 GMT</pubDate>
            <atom:updated>2025-03-05T21:03:13.737Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*ApXTtRJD6NR3e_FU" /><figcaption>.“Di Jacopo da Empoli 1624”</figcaption></figure><h3>The Human Meat-Eating Obsession</h3><p><em>Aww, I love my pets! But holy cow, that grilled chicken on my plate looks so irresistible</em>.</p><p>Globally, 80 billion animals are slaughtered each year for meat. The average will chomp down on 7,000 animals during their lives which breaks down to 11 cows, 27 pigs, 2,400 chickens, 80 turkeys, 30 sheep, and 4,500 fish. Such an interesting fact. You can keep track of live animal slaughter count here — <a href="https://thevegancalculator.com/animal-slaughter/">https://thevegancalculator.com/animal-slaughter/</a></p><p>How did all this start? Why did human beings start eating animals? Let’s dig down this rabbit hole to analyze more on how homo sapiens acquired this immense taste for meat.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/501/0*tb9XB09TYEE3tPi6" /><figcaption>© 2013 <a href="http://www.nature.com/scitable">Nature Education</a> Courtesy of Karen Carr/Smithsonian Institution.</figcaption></figure><h3>Why did our human ancestors eat meat?</h3><p>The main reasons for the hominids to make meat the essential part of their diet is due to the following reasons —</p><p><strong>1.Evolutionary Process</strong></p><ul><li>Without the <strong>abundance of calories</strong> afforded by meat-eating, they maintain, the human brain simply could not have evolved to its current form.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*XG_sPD03Uj4zOorU" /><figcaption>Images by Karen Carr Studios, courtesy of the Smithsonian’s Human Origins Program.</figcaption></figure><ul><li>Meat and marrow are calorie-dense resources with <strong>essential amino acids and micronutrients </strong>(Milton 1999), and aquatic fauna offer resources rich in <strong>nutrients needed for brain growth</strong> (e.g., Broadhurst <em>et al.</em> 2002). Increasing the consumption of animal foods could have allowed hominids to <strong>increase their body size without losing mobility, agility, or sociality </strong>(Milton 1999).</li></ul><p><strong>2. Irreplaceable nutritional value</strong></p><ul><li>In 2009, Claudio Tennie, of the Max Planck Institute for Evolutionary Anthropology, and his colleagues developed a hypothesis that offered a nutritional perspective on the group hunting they had observed in the chimpanzees in Gombe National Park, in Tanzania.</li><li>According to this hypothesis, the <strong>micronutrients gained from meat are so important that even small scraps of meat are worth the very high energy expenditure that cooperative hunting entails</strong>. Important components of meat include not only <strong>vitamins A and K, calcium, sodium, and potassium, but also iron, zinc, vitamin B6, and vitamin B12;</strong> the latter, although necessary for a balanced primate diet, is present only in small quantities in plants.</li></ul><p><strong>3. Ease of consumption</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*BP8tsEJ1LGMIvujD" /><figcaption>Photo issued by Ruth Blasco of marrow inside a metapodial bone after six weeks of storage, as researchers have found evidence that humans living in Qesem Cave in Israel between 420,000 and 200,000 years ago stored long bones so they could tuck into the marrow later. Photo credit: Ruth Blasco/PA Wire</figcaption></figure><ul><li>The pre-human diet majorly started with<strong> fruits and vegetables</strong> (not calorie-dense). A better alternative to this was so-called<strong> underground storage organs</strong> (OSU — root foods like beets, yam, and potatoes) which have better nutritional value. However, they are terrible in taste and hard to chew when it’s raw. This is where<strong> meat </strong>stepped — and ran and scurried — in to save the day. <strong>Prey that has been killed and then prepared either by slicing, pounding, or flaking provides a much more calorie-rich meal with much less chewing than root foods do, boosting nutrient levels overall</strong>.</li><li><strong>On average, they found that it required from 39% to 46% less force to chew and swallow processed meat than processed root foods.</strong></li></ul><p><strong><em>Meat is what made us the most evolved form of homo sapiens.</em></strong></p><p>According to Harvard University evolutionary biologists Katherine Zink and Daniel Lieberman, the authors of the <em>Nature</em> paper, concluded a diet that was 1/3rd <strong>animal protein</strong> and 2/3rd OSUs would have saved early humans about two million chews per year — a 13% reduction — meaning a commensurate saving in time and calorie-burning effort just to get dinner down. <strong>This is the real reason why our human ancestors “had no choice” but to make meat an essential part of their diet.</strong></p><h3>Why does the most evolved form of human beings continue meat consumption?</h3><p>The meat was clearly pivotal in the evolution of the human brain, but that doesn’t <strong>mean that meat is still an irreplaceable part of the modern human diet. </strong>Marta Zaraska (Author of <a href="https://www.amazon.com/Meathooked-History-Science-2-5-Million-Year-Obsession/dp/0465036627">Meathooked</a>) says any calorie-dense food would have had the same effect on our ancient evolving brains — “it could have been peanut butter” — but that meat happened to be available.</p><blockquote>Humans continue to eat meat because we like it, not because we need it.</blockquote><p>For the most part, people globally consume meat for the following reasons.</p><ol><li><strong>Rich source of protein and nutrition</strong></li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/440/0*cyBKNuf4weL0-Q6h.jpg" /><figcaption>Photo by USDA Nutrition Data</figcaption></figure><p>Meat is a complete, high-quality protein containing all of the amino acids required for your body. It provides 25 grams of proteins per 3 ounces than dairy (8 grams per cup), eggs (6 grams each), legumes (12 grams per ¾ cup), vegetables, or nuts (2 to 5 grams per serving). It is the natural source of Vitamin B12. It supports overall optimal health.</p><p><strong>2. Ultimate Craving</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/502/0*CSouDrP7qJheBRzx" /><figcaption>Photo by Huffpost.com</figcaption></figure><p>The Mekeo tribe of New Guinea say that hunger for plant-foods comes from the abdomen, while hunger for <strong>meat starts in the throat.</strong> One thing common among all the animals such as roaches, minks, cats, and humans is the<strong> innate craving for protein</strong>.</p><p>According to the <a href="http://sydney.edu.au/news/84.html?newsstoryid=7975">protein leverage hypothesis</a>, proposed by Stephen Simpson and David Raubenheimer from the University of Sydney, Australia, animals strive to meet a fixed protein target. <strong>Basically, if the amount of calories from protein in the diet falls below about 15 %, we start to crave protein-dense foods</strong>.</p><p><strong>3. The irresistible taste</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*_U-menQ_IOUrxYDO" /><figcaption>Photo by <a href="http://www.thoseshirts.com/mur.html">Thoseshirts.com</a></figcaption></figure><p><strong>Why exactly is meat so tasty?</strong> The answer, according to scientists, lies in meat’s unique<strong> mixture of fat and </strong><a href="https://en.wikipedia.org/wiki/Umami"><strong>umami</strong></a>, spiced up in a process called the <a href="http://modernistcuisine.com/2013/03/the-maillard-reaction/">Maillard reaction </a>— the browning that happens when we cook a piece of meat. “These are powerful stimuli to humans,” says Paul Breslin, a nutritional sciences professor at Rutgers University.</p><p>As much as <strong>95 % of what we think of as meat’s taste is actually its aroma</strong>, according to a Barb Stuckey, author of <a href="http://www.amazon.com/gp/product/1439190747?ie=UTF8&amp;camp=1789&amp;creativeASIN=1439190747&amp;linkCode=xm2&amp;tag=washingtonpost-20">“Taste: Surprising Stories and Science About Why Food Tastes Good</a>.”</p><p><strong>4. Acquired meat-eating behavior since childhood</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/201/0*Etq4J0TjBbYQkDc9" /><figcaption>Photo by <a href="https://kuaibao.qq.com/">https://kuaibao.qq.com/</a></figcaption></figure><p>Dietary quality is influenced by practices such as <strong>eating breakfast, family meals, and fast food consumption</strong>. Both dietary quality and meal practices are linked with <strong>sociodemographic characteristics</strong>. It is extremely likely that a kid who has been brought up by eating non-vegetarian food would continue to do so throughout his entire lifetime. Convincing people to make changes in their diet is not a simple task since they have been following different practices since their childhood. However, education plays a huge role in assisting the shift in dietary habits of people to prevent the animal killing.</p><h3>What do the animals experience while human beings continue meat consumption?</h3><p><em>Disclaimer: Graphic content consists of human cruelty and animal slaughtering</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/240/0*sBQ-XE-rOYEz1WNx" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*5QrUBmy2eLWfNDl6" /><figcaption>Photo by <a href="https://www.kinderworld.org/">https://www.kinderworld.org/</a> and <a href="https://abundantpermaculture.com/">https://abundantpermaculture.com/</a></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/278/0*Q38SkAaME8MTXrqI" /><figcaption>Photo by <a href="https://animalequality.org/">https://animalequality.org/</a></figcaption></figure><p>Guess the common emotion among the above images — “Pain”.</p><p>Yes, that’s 100% right! They are all in PAIN. However, let’s understand the techniques in which the<strong> animals are prepared to be our yummy dinner</strong>.</p><p>The most widely used technique of killing animals adopts the principle of the two-stage process of making the animals <strong>unconscious and then killing</strong> them for the non-ritual slaughter of animals —</p><ul><li><strong>Electrical Stunning</strong> — <strong>Current is applied either across the brain or the heart</strong> to make the animal unconscious before they are killed. This is basically, <a href="https://www.hsa.org.uk/electrical-stunning/electrocution#:~:text=Electrocution,oxygen%20and%20will%20rapidly%20die."><strong>electrocuting</strong></a><strong> </strong>the animal.</li><li><strong>Asphyxiation</strong> — Animals are <strong>put in a chamber filled with 80% to 90% CO2</strong> which helps in losing their consciousness.</li><li><strong>Captive Bolt Pistol Strike</strong> — Animals are <strong>forcefully struck with a </strong><a href="https://en.wikipedia.org/wiki/Captive_bolt_pistol"><strong>captive bolt pistol</strong></a><strong> on their forehead.</strong></li><li><strong>Exsanguination</strong> — The animal’s <strong>throat is cut or a chest stick is inserted close to the heart</strong> so the main arteries and veins are cut and allowed to bleed.</li></ul><p>Now, let’s all imagine the exact same situations described above happening to either <strong>human beings</strong> (any race, any ethnicity, any gender, any age group, anybody at all) or our <strong>beloved dogs, cats, or so-called “pets” at home</strong>; How do we feel? What images do we see? Do we have the courage to think about this? Would we all still continue leading our lives normally without any protests?</p><h3>To be continued.</h3><p>References</p><ul><li><a href="https://thevegancalculator.com/animal-slaughter/">Animals Slaughter Kill Counter- The Vegan Web Designer</a></li><li><a href="https://www.usatoday.com/story/news/nation-now/2015/03/11/meat-eaters-animals-lifetime/70136010/">Holy cow! Meat eaters consume 7,000 animals in lifetime</a></li><li><a href="https://www.americanscientist.org/article/meat-eating-among-the-earliest-humans">Meat-Eating Among the Earliest Humans</a></li><li><a href="https://time.com/4252373/meat-eating-veganism-evolution/">Sorry Vegans: Here&#39;s How Meat-Eating Made Us Human</a></li><li><a href="https://www.nature.com/scitable/knowledge/library/evidence-for-meat-eating-by-early-humans-103874273">Evidence for Meat-Eating by Early Humans</a></li><li><a href="https://mymeatup.org/content/nutrition">Seven Nutrition Facts About Meat That Might Surprise You</a></li><li><a href="https://www.huffpost.com/entry/meat-hunger-is-real-for-some-people-not-you_b_9632646">Meat Hunger Is Real For Some People, But You&#39;re Probably Not One Of Them</a></li><li><a href="https://www.washingtonpost.com/national/health-science/what-makes-a-hamburger-and-other-cooked-meat-so-enticing-to-humans/2013/08/12/8f8e1d72-ff73-11e2-9711-3708310f6f4d_story.html">What makes a hamburger and other cooked meat so enticing to humans?</a></li><li><a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6024598/">Factors Influencing Children&#39;s Eating Behaviours</a></li><li><a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6165406/">Selected Psychological Aspects of Meat Consumption-A Short Review</a></li><li><a href="https://en.wikipedia.org/wiki/Animal_slaughter">Animal slaughter</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d3406c24837" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Facebook PyTorch Udacity Scholarship challenge]]></title>
            <link>https://medium.datadriveninvestor.com/facebook-pytorch-udacity-scholarship-challenge-a269fe682805?source=rss-98d422ea8e5d------2</link>
            <guid isPermaLink="false">https://medium.com/p/a269fe682805</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[udacity]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[pytorchudacityscholar]]></category>
            <dc:creator><![CDATA[Anupama Garani]]></dc:creator>
            <pubDate>Sun, 20 Jan 2019 09:48:00 GMT</pubDate>
            <atom:updated>2019-01-22T14:00:23.539Z</atom:updated>
            <content:encoded><![CDATA[<figure><a href="http://www.track.datadriveninvestor.com/1B9E"><img alt="" src="https://cdn-images-1.medium.com/max/700/0*McssR8tiwSC0bQW8" /></a></figure><p>This blog post is about my journey through my second scholarship from Facebook and <a href="https://www.udacity.com/facebook-pytorch-scholarship">Udacity </a>to learn the basics of neural network and build projects in deep learning using Facebook’s <a href="https://pytorch.org/">PyTorch </a>framework.</p><p>Everyone knows how effective and rewarding it feels when you learn something with a community of peers who are working on the same curriculum as you. But I’ve to admit it, I had never thought e-learning would also be this much fun. All thanks to Udacity! Udacity has one of the best community managers and slack team for progressing through the coursework without any hindrance to any of the students.</p><p>But, firstly how did I even get selected to the scholarship challenge might be your question! So, one fine snowy (I stay in NY folks, and yes the struggle is real) day I bumped across this mail in my inbox announcing scholarship from Facebook and Udacity for data nerds all over the world and I knew this was my chance to take a jab at structured learning for AI using PyTorch. (Internet is so overwhelming to start with). And YEESS, I did make it to the scholarship challenge..like one among the 10,000 amazing people all over the world.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/483/1*EfDQU_uuRduae2Th1nI_FA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0rTKTnJNS9KQMbN6.png" /><figcaption>Yay, made it into the first Phase!</figcaption></figure><p>Well, it wasn’t too bad, until the Slack forum was overpouring with thousands of message from people. I should say, it looked very abused with lots of channels — motivational, live-help, study-buddies, regional groups, LinkedIn-connections, fun torch and more! Slack was the go-to place for any discussions and queries you had. The average estimated wait time for your reply would be close to 0.5 minutes for any question from my observation and experience.</p><p>Let’s cut the chase and get to the crux of the program. So, the coursework offered a very methodical and organized way for users to absorb the concepts of AI and deep learning, given the math behind it is extremely painful to master within a short period of time. It was structured the following way.</p><h4><strong>Introduction to Neural Networks</strong></h4><p>Introduction to the concepts of NN along with the purpose of its existence (sophisticated classification) and the pieces which make NN such as <em>perceptron algorithm, gradient descent algorithms, entropy, dropout, regularization, and vanishing gradient issue, local and global minima.</em> This was by far the most visited chapter by peers and myself to get hang of the jargons for the advanced neural nets. Oh yes, we also got to code the 2-layer NN and gradient descent algorithm from scratch.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/347/1*Y7bXC2dhLE0zYlpMUHEuQg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/367/1*QFVtUnylvdbiznJcE3OJhg.png" /><figcaption>Gradient Descent Algorithm</figcaption></figure><h4>Creation and future of PyTorch Framework by the creator, Soumith Chintala</h4><p>The creator himself walking us through the details of how PyTorch came into existence from research labs to deployment and how it’s been a success in modeling Neural Nets and the major improvements to the framework.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZyegdA0xvDaYg26CAdsIlg.png" /><figcaption>Discussion with Soumith Chintala</figcaption></figure><h4>Convolutional Neural Network</h4><p>This was my most favorite part of the course coz it deals with a lot of image datasets, and it&#39;s so cool to play around with them to create something creative out of it. Neural nets have the ability to identify the image which you throw at it and not just that, it can detect edges in an image, photoshop, style your image and the best — It can CREATE most amazing images! Some of the programs which I wrote are on Github. <a href="https://github.com/anusky95/PytorchPrograms/blob/master/README.md">https://github.com/anusky95/PytorchPrograms/blob/master/README.md</a> and <a href="https://github.com/anusky95/Convolutional-Neural-Network">https://github.com/anusky95/Convolutional-Neural-Network</a> (Promise to push ‘em all on git soon.)</p><h4>Style Transfer</h4><p>This was a very interesting CNN implementation where the style of one image can be applied to the other content image and the resultant image contains the style of the second image on the first one. This is achieved by extract the feature maps and the overall content representation of the first image and applying the textures, colors, and style of the second image on the content image. An example of this is shown below in the style transformation. It was a lot of fun doing this activity (despite the time it took to train the models on Google Colab)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5mH4L2DjaAxq0VyQnm2JeA.jpeg" /><figcaption>Style image: My canvas poster paintings</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ipYWdz5-mIgEV0u6vbpVag.png" /><figcaption>Content Image (View from RIT shuttle stop) &amp; Result Image: Looks like an art work</figcaption></figure><h4>Recurrent Neural Network</h4><p>While CNN’s are used for images, RNNs are used for learning complex patterns from sequences of unstructured data such as Natural Language Processing. LSTM’s (Long Short Term Memory) is an advanced form of RNN as it can switch from remembering things from long term memory and short term memory and thus outperforms RNN. Learning LSTM took quite a while considering the role of the various gates.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NEHHTH7PAtuBKDH7CNgLAQ.png" /><figcaption>And this is how the architecture of LSTM looks like</figcaption></figure><h4>Final Project</h4><p>The final project was to build an image classifier for 102 different species of flowers which can be downloaded from <a href="http://www.robots.ox.ac.uk/~vgg/data/flowers/102/index.html">here</a>. Since the dataset was large, it takes a lot of time to model on CPU and trained the model on Google Colab. A <a href="https://medium.com/deep-learning-turkey/google-colab-free-gpu-tutorial-e113627b9f5d">tutorial </a>is worth a shot to avoid wasting a lot of time on setting it up. Udacity provided us with <em>training and validation</em> set only and had us upload the model on the workspace which would then test the images on the hidden test flower images. If the accuracy was at least 20%, the model was project was considered to be complete. However, a lot of us strived to achieve very good accuracy and the <strong>extremely </strong>( I wish I could emphasize enough) helpful community of 10k peers on slack played a huge role in this. Here is the github link to the final project — <a href="https://github.com/anusky95/Udacity-PyTorch-Challenge-Scholarship-Project/blob/master/README.md">https://github.com/anusky95/Udacity-PyTorch-Challenge-Scholarship-Project/</a></p><p>Below are some of the wonderful moments we had on slack!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7v05jv2wSNqPVpdUvs69ZA.png" /><figcaption><strong>This is how the final project submission day looks like!</strong></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-i_Q3ORGepAngagT38CGdA.png" /><figcaption><strong>We had Study Jams to help us rock and roll!</strong></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DIO5zjecblkFhhY7owkMSA.png" /><figcaption><strong>Announcements for side projects and fun activities</strong></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*xIRP5SVXr-7I60UKDj4soQ.png" /><figcaption>Spot me!!!! ;)</figcaption></figure><h4>Side Project (BROWNIE POINTS)</h4><p>Okay, this last bit was totally us on what we would do using PyTorch Framework. So, I decided to have some fun with RNN. I built a name-ninja generator which generates random names based on the thousands of names you feed to the model using Char-based RNN. More about the project in the next post :)</p><p>Thanks for reading the long post! ❤ :)</p><p>Follow me for more posts and articles on Data Science.</p><p>LinkedIn — <a href="https://www.linkedin.com/in/anupama-garani/">https://www.linkedin.com/in/anupama-garani/</a></p><p>Twitter — <a href="https://twitter.com/agarani95">https://twitter.com/agarani95</a></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fb2a0d6%3Fas_embed%3Dtrue&amp;dntp=1&amp;display_name=Upscribe&amp;url=https%3A%2F%2Fupscri.be%2Fb2a0d6%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href">https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a269fe682805" width="1" height="1" alt=""><hr><p><a href="https://medium.datadriveninvestor.com/facebook-pytorch-udacity-scholarship-challenge-a269fe682805">Facebook PyTorch Udacity Scholarship challenge</a> was originally published in <a href="https://medium.datadriveninvestor.com/">DataDrivenInvestor</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Scrub your data until it speaks out stories!]]></title>
            <link>https://medium.com/udacity-bertelsmann-scholarship-blog/scrub-your-data-until-it-speaks-out-stories-355ad7768eac?source=rss-98d422ea8e5d------2</link>
            <guid isPermaLink="false">https://medium.com/p/355ad7768eac</guid>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Anupama Garani]]></dc:creator>
            <pubDate>Thu, 02 Aug 2018 10:27:09 GMT</pubDate>
            <atom:updated>2018-08-02T10:27:09.593Z</atom:updated>
            <content:encoded><![CDATA[<iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fgiphy.com%2Fembed%2F3o7TKqTOqevMiJAwU0%2Ftwitter%2Fiframe&amp;url=https%3A%2F%2Fmedia.giphy.com%2Fmedia%2F3o7TKqTOqevMiJAwU0%2Fgiphy.gif&amp;image=https%3A%2F%2Fi.giphy.com%2Fmedia%2F3o7TKqTOqevMiJAwU0%2Fgiphy.gif&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=giphy" width="435" height="435" frameborder="0" scrolling="no"><a href="https://medium.com/media/9f9f05e4984ab16bd5f5030fa928e7b7/href">https://medium.com/media/9f9f05e4984ab16bd5f5030fa928e7b7/href</a></iframe><p>According to a big data article published in Forbes by Press (2016), data scientists spend 60% of their time on pre-processing and organizing data. Well, that’s a bit too much! Aren’t they supposed to be working on researching and developing new machine learning based algorithms? So, why waste time on pre-processing and what exactly does it mean?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/0*Pi9SlNhaXnPSe-Q0" /><figcaption><a href="https://goo.gl/T42f6h">https://goo.gl/T42f6h</a></figcaption></figure><p>It simply means — <strong><em>scrubbing the dirt off the data</em></strong> and<strong> <em>altering</em> </strong>it to prepare for better <strong><em>visualization</em></strong> and better <strong><em>readability</em></strong> to derive meaning from the data.</p><p>Ahh! Sounds simple — erm, but, what kind of dirt is any data set associated with? How does it appear like? How do we know it’s dirt and it needs to be removed? And what kind of changes needs to be made?</p><p>Well to answer all these questions, let’s dive a bit deeper into data; traditionally most of the data scientists follow these steps in pre-processing (That’s what wikipedia says!)</p><ul><li><strong><em>Data Cleaning</em></strong></li><li><strong><em>Instance selection</em></strong></li><li><strong><em>Normalization</em></strong></li><li><strong><em>Transformation Feature extraction</em></strong></li></ul><p>That’s a lot of buzzwords! I thought it was just cleaning the data and making it look good. Let’s take baby steps and get to the first step and explore what is <strong><em>data cleaning</em></strong> in this blog.</p><p>Identify the 3I’s in the data and keep an eye on them!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/0*5UAkV3-zwGqu_qIb" /><figcaption>Types of dirt that needs to be scrubbed off the data</figcaption></figure><p>For instance, consider the below table where the survey answer for the question “How would you describe your life” is<strong> incomplete and inconsistent</strong> which does not convey any meaning or useful inference. In the second column “Age of your grandmother” has a response of 10 years old which is obviously <strong>inaccurate and not right</strong>. This leads to inaccurate analysis since the mean of the age data column will be affected by the wrong age. In the third column, annual pay rate of an employee in a state does not contribute to analysis of the diseases in the states and hence such data can be ignored since they are irrelevant.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/560/0*--iB9pYGwdC9CheI" /><figcaption>An instance of the 3 I’s lurking in the dataset</figcaption></figure><p>Let us consider an actual data set and identify the dirt settled on them</p><p>Dataset Source : <a href="https://think.cs.vt.edu/corgis/csv/music/music.html">https://think.cs.vt.edu/corgis/csv/music/music.html</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/527/1*dTmRazwHBOgXO6YIWodswQ.png" /><figcaption>Loading the data</figcaption></figure><p>We shall first explore the attributes of the music data set by printing all the columns in the data frame.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/542/1*QZrHdI4HLKPGEIHRosXJDg.png" /><figcaption>Checking for Null values</figcaption></figure><p>Python has this incredible power of spotting the presence or absence of NULL values in the data set.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/555/1*09ge0AyQpuGSssikUfR2sQ.png" /><figcaption>Printing the count of the null values</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/594/1*J8V3LlO_3nCUDWuxh0bgvw.png" /><figcaption>Null value distribution in the music dataset</figcaption></figure><ol><li><strong><em>Missing Values</em></strong></li></ol><p>As we can observe, the attribute <strong>artist_mbtags</strong> has the most number of null values in the dataset when compared to the other attributes. Around <strong>63%</strong> of the values of the <strong>artist_mbtags </strong>has NaN value which accounts to more than half of the dataset being null valued for that attribute. We can handle this by eliminating the attribute which would be the best method of action.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/654/1*SpVYXiOofjBuKIivRtHNeA.png" /><figcaption>Attributes which contains null values</figcaption></figure><p>Let’s take one of the attribute given and dig deep and see what kind of values it contains.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/663/1*Kk4gKGm7RGPOhg1GzsEHMQ.png" /><figcaption>Selection of a single attribute for checking null value</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/545/1*xI7Ch7NPVUbDuqvJJooPxA.png" /><figcaption>Statistics for the song hotness attribute</figcaption></figure><p>As we can see from the previous null value distribution table, the song hotness attribute has around 43% null values. This is less than half of the data set and can be replaced with either mean, median or mode or zero’s. For better understanding of the data, it is always a good practice to perform exploratory data analysis.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5aEwYM1BdoA0mRWFz3o-pw.jpeg" /><figcaption>Song hotness measure of each song (10,000 values)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4jR33JqNu4TCdQRLZC72IA.jpeg" /><figcaption>Song hotness measure of songs (sampled)</figcaption></figure><p>As we can see in the second graph, the drilled down sampled version of the song hotness of each album clearly shows the presence of NULL values.</p><p><strong><em>2. Inaccurate values</em></strong></p><p>Inaccurate text values are relatively easier to notice in the excel sheet for which corrective actions can be implemented using python scripting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/605/0*O-odE06kZtAc9GSb" /><figcaption>Inaccurate song name with special characters</figcaption></figure><p>NLP (Natural Language Processing) can be implemented on any text related data which makes the analysis simpler. But text data like the one shown in above figure can be handled by regular python scripting.</p><p><strong><em>3. Irrelevant attributes</em></strong></p><p>The main idea behind using the data set is to understand and analyze the factors affecting the popularity of music created by the artists . This scenario, does not demand the need of the <strong>release.id </strong>attribute since the relationship between the popularity of the music does not depend on the release.id nor can any conclusions be drawn from that.</p><p>That’s it for now folks, more to it in my next article! Enjoyed writing my first article on Medium :) Hope you got something out of it too!</p><p>Thanks to Udacity Bertelsmann’s program and to women_of_code for giving me a chance to express my thoughts!</p><p>If you enjoyed the content, please leave a 👏</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fgiphy.com%2Fembed%2F9Jcw5pUQlgQLe5NonJ%2Ftwitter%2Fiframe&amp;url=https%3A%2F%2Fmedia.giphy.com%2Fmedia%2F9Jcw5pUQlgQLe5NonJ%2Fsource.gif&amp;image=https%3A%2F%2Fmedia.giphy.com%2Fmedia%2F9Jcw5pUQlgQLe5NonJ%2Fgiphy-downsized-large.gif&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=giphy" width="435" height="435" frameborder="0" scrolling="no"><a href="https://medium.com/media/573c5a971b93cfc1671a4820cb75e525/href">https://medium.com/media/573c5a971b93cfc1671a4820cb75e525/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=355ad7768eac" width="1" height="1" alt=""><hr><p><a href="https://medium.com/udacity-bertelsmann-scholarship-blog/scrub-your-data-until-it-speaks-out-stories-355ad7768eac">Scrub your data until it speaks out stories!</a> was originally published in <a href="https://medium.com/udacity-bertelsmann-scholarship-blog">Udacity Bertelsmann Data Science Scholarship 2018/19 Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>