Lost in the AI Hype, I Started Small — DeepSeek Blog | Neura Market
    Neura MarketNeura Market/DeepSeek
    ChatGPTChatGPTClaudeClaudeGeminiGeminiCursorCursorGrokGrokPerplexityPerplexityDeepSeekDeepSeek
    CoPilotCoPilotStable DiffusionStable DiffusionMidjourneyMidjourney
    View All Directories
    OverviewRulesPromptsMCPsAgentsBlogVideosGuidesCoursesCommunityTrendingGenerate
    DeepSeekBlogLost in the AI Hype, I Started Small
    Back to Blog
    Lost in the AI Hype, I Started Small
    ai

    Lost in the AI Hype, I Started Small

    Rohini Gaonkar April 15, 2026
    0 views

    And it helped me get back into tech without drowning TL;DR at the end Coming back to...

    ## And it helped me get back into tech without drowning > [TL;DR](#tldr) at the end Coming back to tech after a (maternity) break is a strange feeling. You’re excited but also unsure where to begin. There are new tools, new terminologies and new way of doing things we did for decades. But I didn’t try to figure everything out at one go, I just picked one small thing. For me, that “one small thing” was finally building my portfolio collection. Over the years, my content - blog posts, YouTube videos, conference talks, GitHub repos, social posts all of them scattered across dev.to, GitHub, YouTube, Instagram, LinkedIn, and AWS channels - has been scattered across a dozen different platforms. DEV.to, community.aws, YouTube, GitHub (two accounts!), LinkedIn, SlideShare... you name it. 🙇‍♀️ More than 80 pieces of content, scattered across platforms since 2015!!! And honestly? Maintaining my existing site [rohinigaonkar.com](https://rohinigaonkar.com) felt harder than starting from scratch. I wanted something simpler, a lightweight site I can update by editing a single file, push to GitHub, and it's live. Easy to navigate, easy to maintain. No fluff. I built this as my first project back from maternity leave, and I did it with [Kiro](https://kiro.dev/?trk=44b16281-e090-49b6-97d8-f1cea54d9e87&sc_channel=el), an AI-powered IDE from AWS that I'd never used before. Two firsts at once. It turned out to be the perfect re-entry project. ## Building It with Kiro: My First Impressions This portfolio had been on my mental to-do list forever, so the timing felt right. And rather than spinning up a complex stack to shake off the rust, I decided to keep it simple and lean on an AI coding assistant to help me get back into the flow. Here's what stood out about the experience, starting with the simplest features and building up. > > Last one might surprise you!!! 🤯 > ## 1. Chat-Driven Development The entire project was built through conversation. ![Screenshot: Kiro Kiro-IDE.png](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/cc8xwf0zsvgryqcqnkor.png) I described what I wanted, *"I have this website where I collect my content shared across multiple social media websites. it is one true place where any tech content I posted on the web be it dev.to, github, youtube, instagram, and any aws first party channels, all of this to be collected as a timeline. can we build something that can be refreshed on demand and build this portfolio. make it professional looking. ask more intelligent questions as we go."*, and Kiro asked clarifying questions even before writing a single line of code. It asked about static vs dynamic, hosting preferences, design vibe, and data source approach. It even asked me about my identity. That back-and-forth shaped the architecture before any code was generated. ![Screenshot: Initial chat conversation with Kiro asking clarifying questions Kiro-asking-questions-before-coding](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/fyxbyiv6ioht034zklax.png) Once I provided response to all the questions, it help build the initial structure and also walked me through it. ![Screenshot: Initial build initial-build-2](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/gtjfxdcazqqqknz26je2.png) Notice how after every conversation, it shows how many credits each prompt consumes, in real-time. That was nice! There is a **spec-driven development mode** as well, which I would be testing for something more complex than this static website. ![Screenshot: Kiro's spec-driven development mode spec-driven-development](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/v9qhqndj2v4fd9wj9dtd.png) ## 2. Web Search Kiro searched the web to find my existing content across platforms. It looked up my dev.to profile, GitHub repos, community.aws presence, YouTube channel, and even my current website at rohinigaonkar.com. This gave it real context about who I am and what content already exists, so the portfolio wasn't built with placeholder data, it was seeded with my actual content from day one. ![Screenshot: Kiro web search results finding your profiles web-search](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/jij1zq0izq98o0zssb48.png) ## 3. Explore Real-Time File Changes As Kiro generated and edited the files, I could see every change happening in real time through the explorer. Either click the **"Follow"** option or click on the little **"diff button"** highlighted with yellow square below. For example, it created `index.html` or `content.js`, I could immediately open them, review the code, and see the diffs. When it later modified `content.js` to add YouTube videos or reclassify talks vs videos, I could see exactly what changed and why. ![Screenshot: Explorer view showing file changes / diff view Follow-file-changes](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/klirbnx1pxu8f7h0ld6s.png) ## 4. Trusting Frequently Run Commands Kiro ran shell commands like `curl` to hit APIs and extract data. It asked my permission to **run it once or add it to trusted list of commands.** ![Screenshot: Autopilot mode with command execution run-or-trust-command](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/hj5obnlcn3c8v23gfe1b.png) I also liked how it provided my levels of trust, like I can just execute this particular command or partial or the base command. ![Screenshot: Autopilot mode with command execution trust-levels](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/zpfe0xsmllwnoietp1fp.png) In autopilot mode, I could trust these commands to execute without approving each one individually. This was especially useful during the YouTube oEmbed batch processing, where Kiro ran 16 consecutive curl commands to fetch video titles, approving each one manually would have been tedious. ## 5. Iterative Refinement Through Conversation The project evolved through multiple rounds of feedback: - I pointed out that "talks" should only mean conference/meetup presentations, not YouTube tutorial videos - Kiro reclassified everything accordingly - I noted that some AWS "talks" were actually just YouTube embeds on my website - Kiro dug into the pages, extracted the real YouTube URLs, and recategorized them as videos - I shared my personal GitHub profile separately from my work one - Kiro pulled repos from both and updated the refresh script to handle multiple accounts Each round of feedback made the portfolio more accurate without starting over or deleting some other important information. ![Screenshot: Chat showing iterative refinement iterative-refinement](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/q1mjsvtt5g6etfeh8yfd.png) ## 6. The YouTube Challenge : Hitting Walls and Finding Workarounds This was the most interesting part as YouTube is heavily locked down: - **Direct fetch failed** - `webFetch` on youtube.com returned empty content - **Rendered mode failed** - returned only JavaScript bootstrap code, no actual page content - **Search was noisy** - web searches for my videos returned generic results, not my specific content - **RSS feeds blocked** - YouTube's channel RSS wasn't accessible either > > **But Kiro didn't give up!!! 💜 ** > It found workarounds on its own: - **oEmbed API** - Kiro discovered that YouTube's oEmbed endpoint (`youtube.com/oembed?url=...`) returns video titles as JSON, and used `curl` to call it directly. This became the reliable method for all 16 videos I shared. - **Squarespace page parsing** - For videos embedded on my website, Kiro parsed the raw HTML to extract YouTube video IDs from Squarespace's embed block JSON (double HTML-unescaping the content to find URLs like `youtube.com%2Fembed%2Fi0zQpJPfSdU`). - **Thumbnail URL extraction** - It even tried extracting video IDs from `ytimg.com/vi/VIDEO_ID/hqdefault.jpg` thumbnail patterns as a fallback. The iterative problem-solving here like trying one approach, hitting a wall, pivoting to another, felt very much like how a developer would debug a scraping problem. I loved how Kiro told me what it tried, it failed and it was going to try something else. We also worked together for a process that was a good compromise for both of us. Maybe in future I will have an agent to simplify this, but for now this solves my purpose! ![Screenshot: Kiro trying different approaches to fetch YouTube data youtube-is-blocked-1](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/rzq0zn9f7x8wjfnf9t6j.png) ![Screenshot: Kiro trying different approaches to fetch YouTube data youtube-is-blocked-2](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/yth8gchj9tap5j1wndv6.png) ![Screenshot: Kiro trying different approaches to fetch YouTube data youtube-is-blocked-3](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/bwfh7246d1c4rpxzt1vh.png) ## The Result A live static portfolio at [rohinigaonkar.github.io](https://rohinigaonkar.github.io) with ~80 entries spanning 2015–2025, filterable by type (blogs, videos, repos, talks, social), searchable, and refreshable on demand via a Node.js script that pulls from GitHub and dev.to APIs. All built through conversation in a single Kiro session. ![Screenshot: Final portfolio site summary quick-recap-summary](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/xfkgvhvnhe21he0zwr9e.png) ![Screenshot: Final portfolio site final-website](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/j3tydhj0wrh2llspzsss.png) ## What's Next I'll keep adding content as I publish it. The refresh script makes the API-sourced stuff automatic, and the manual entries take about 30 seconds each. I might add dark mode at some point, and maybe an RSS feed. But right now, the simplicity is the feature. More importantly, this project reminded me that coming back doesn't have to be intimidating. If you're returning from a break and looking for a low-pressure way to get back into coding, I'd recommend picking a passion project and giving Kiro a try. You might surprise yourself with how quickly you get back into the zone. Sign up for an [AWS Builder ID](https://aws.amazon.com/builder/?trk=44b16281-e090-49b6-97d8-f1cea54d9e87&sc_channel=el), install [Kiro IDE](https://kiro.dev/?trk=44b16281-e090-49b6-97d8-f1cea54d9e87&sc_channel=el), and try it for yourself. Be sure to let me know how it goes! And follow along as I explore this world of AI. <a id="tldr">**TL;DR:** </a> Returning to tech after maternity leave felt overwhelming, so I started small by building a portfolio site to consolidate 80+ pieces of content scattered across platforms since 2015. Using AWS Kiro (an AI-powered IDE) for the first time, I built the entire project through conversation—no complex setup needed. Kiro helped with web searches, real-time file changes, iterative refinements, and creative problem-solving (especially when YouTube's APIs were locked down). The result: a live, searchable portfolio at rohinigaonkar.github.io that's easy to maintain. The lesson? Coming back doesn't have to be intimidating—pick one small passion project and let AI tools help you get back into the zone.

    Tags

    aibeginnersproductivityaws

    Comments

    More Blog

    View all
    How I'm using ASTs and Gemini to solve the "Codebase Onboarding" problem 🧠ai

    How I'm using ASTs and Gemini to solve the "Codebase Onboarding" problem 🧠

    Hi everyone! 👋 I’m Tara, a Senior Software Engineer and Consultant. Over the years, I've jumped...

    T
    tworrell
    Local AI Will Save Us All (The Math Says So, Trust Me)ai

    Local AI Will Save Us All (The Math Says So, Trust Me)

    Every few weeks a take goes viral in tech circles making the case for ditching cloud AI and running...

    S
    Sebastian Schürmann
    Building a Replay-Tested Interactive Brokers Client in Gogo

    Building a Replay-Tested Interactive Brokers Client in Go

    I wanted an IBKR library that felt like Go and had testing I could trust. So I wrote one.

    T
    Thomas Marcelis
    Playwright in Pictures: Fully Parallel Modeplaywright

    Playwright in Pictures: Fully Parallel Mode

    Playwright’s fullyParallel mode is often treated as a simple performance switch. In practice, it...

    V
    Vitaliy Potapov
    Designing a CLI for Both Humans and Agentscli

    Designing a CLI for Both Humans and Agents

    Learn how Alpic designed its CLI for both human developers and AI agents — covering tradeoffs like polling, context windows, interactivity, and statelessness.

    J
    Julien Vallini
    Shik — I finally feel joy writing scriptsprogramming

    Shik — I finally feel joy writing scripts

    You know that feeling when the thought in your head is simple and clear — "go through files, find the...

    M
    Maksim Iakovlev

    Stay up to date

    Get the latest DeepSeek prompts, rules, and resources delivered to your inbox weekly.

    Neura Market LogoNeura Market

    Discover the best AI prompts, plugins, and resources for DeepSeek and more.

    Content Types

    • Rules
    • Prompts
    • MCPs
    • Agents
    • Guides

    Platforms

    • ChatGPT Directory
    • Claude Directory
    • Gemini Directory
    • Cursor Directory
    • Grok Directory
    • Perplexity Directory
    • DeepSeek Directory
    • CoPilot Directory
    • Stable Diffusion Directory
    • Midjourney Directory
    • All Directories

    Resources

    • Blog
    • Documentation
    • Help Center
    • Marketplace

    Legal

    • Privacy Policy
    • Terms of Service

    © 2026 Neura Market. All rights reserved.

    |

    Not affiliated with any AI platform vendors.