Thoxt founder Shilpa Ahuja shares the challenges and aha-moments behind building Thoxt, which aims to create a gig economy in the media/blogging industry.
When Shilpa Ahuja set out to build Thoxt – a writer-first publishing platform – she didn’t have a tech background. She also didn’t have a team. What she did have was a decade of editorial experience, a vision for creative expression in the post-AI era, and a willingness to learn programming from scratch. In this interview, she shares how she built a full-stack product, and how turning her editorial instincts into algorithms helped make Thoxt feel like more than just another social media app.
Founder Journey & Learning to Code
You come from a non-tech background – what made you decide to learn programming yourself?
Shilpa: My co-founder (Abhishek Sareen) and I had the idea for Thoxt two years ago, and since none of us had tech skills to build it, we spent months looking for a tech co-founder or agency. Eventually, I realized that I can’t build a tech startup without knowing tech, just like I couldn’t build a blogging business without knowing how to blog. So I decided to fully give in and learn software engineering from scratch.
The idea of Thoxt was a culmination of all these years I spent writing and running a blogging business, it seemed like the way the blogging industry should be heading in the future, it was too good to pass up! So that became my incentive to do the hard work required to learn programming.
How did learning to code in the AI era differ from, say, 5-10 years ago?
Shilpa: It’s very different, part in thanks to Covid making world-class courses and tutorials available online, and part in thanks to AI tools that can be your guide for everything. I’m so glad to have access to everything so I could learn right at home while raising toddlers and running my business, otherwise it’d have been impossible!
AI is like having a 24/7 tutor that can help you understand concepts whenever you want, so if you really like learning, you can understand tradeoffs between different technologies, libraries, frameworks, languages, which immensely speeds up the learning process and makes it fun. Of course, there’s still hard work and time commitment required.
5-10 years ago I’d have to spend hours looking up forums, hoping that someone has asked and answered my questions; reading documentation for everything and maybe giving up on certain technologies just because of how time-consuming or hard it’d be to understand them. Although, 5-10 years ago nobody thought you could just wake up one day and decide to become a programmer! It was unthinkable.
What was the biggest mental shift going from media founder to tech builder?
Shilpa: The biggest difference between tech and media is the eagerness in adoption of new technologies. Legacy media companies have always led the media industry, so for new media startups, there are already set standards to adhere to whether it’s about content quality, topics, aesthetics, sales or distribution. Being a media founder always felt like playing a game of catch-up that you can never win.
In tech, however, new startups emerge as leaders every few years or so. So it’s very fast-paced, unexpected, and the focus is on innovation. For a creative person like myself, it’s great to realize that I can just dream up my own future and build it without worrying about competitors, and all I need to do is work hard and keep at it.
Being a tech builder is all about keeping up with the latest tech, learning from your users and iterating fast, even if it means changing main features or UI – as long as it aligns with your overall vision. Tech startups are all about thinking and building new, better solutions for users, whereas media companies are all about preserving the good things that are already there.
How did you balance learning to code with parenting and building a startup?
Shilpa: Wow, it’s been a hell of a journey; it was so hard I’d rather not even think about it or it’ll make me cry! Raising twin toddlers is already a task that can tire out even the most hard-working mom to the degree they can’t even work full time, let alone learn something as hard as software engineering.
Maybe a balancing act is got me through – my CS classes were my escape from the parenting madness, and spending time with my kids was my escape from when my code won’t work!
I’ve always loved learning new things, and finishing computer science courses was so fulfilling because it proved that you can do anything at any age! Building a tech startup is just like raising a kid – it can be unpredictable, extremely hard and also worth it. I’m glad I had help along the way!
What do you think more non-tech founders misunderstand about tech?
Shilpa: The biggest misunderstanding is that the ideas are worth everything, and with the right idea, you can get it built from a team or employee, while you reap the benefits! The reality is vastly different – there’s a reason all the best tech startups have a tech founder who got their hands dirty and built the tech themselves from the ground up. No one else can do it for you, not even AI!
The thing is, media glamorizes tech and AI startups as overnight successes, and the news tends to ignore the years of hard work that went behind every successful one. Non-tech founders love such news because it makes it seem that their ideas are worth a lot (no one should steal it). The truth is it’s 1% ideation and 99% perspiration.
Product & Architecture
What’s the tech stack behind Thoxt, and why did you choose it?
Shilpa: The back-end is Python, Django and MySQL. Front-end is vanilla JavaScript, jQuery, CSS, some React for the games and all the basic stuff. I chose Python because it has lots of libraries for building neural networks, in the coming days we’re working on a mood-based topic suggestion tool, ad and recommendation engines, all of which will be easy to prototype with it.
On the AI front we use integrated language queries, power by OpenAI and LLama models to automate SEO, generate content meta data, train a topic suggesting tool and enhance search.
We had a conversation last week about Vite. Why did you uninstall it – and what did you learn from that experience?
In the process of building and optimizing, you try many different libraries and some of it works for your use case, some doesn’t. I initially chose Vite because it promised a modern dev experience , fast builds, hot module replacement. Thoxt is a complex multi-app Django platform with dynamic JS needs, and I use Django’s built-in staticfiles manifest system that clashes with Vite, so things started getting messy.
The best way to optimize Thoxt is that I load different JavaScript files conditionally, only when a user clicks something or navigates to a certain page. Vite doesn’t offer that level of modular control. CSS was also an issue, importing CSS dynamically with JS wasn’t smooth, and paths kept breaking across apps. Sometimes, using a framework or library feels like you’re fighting the tool and you’re better off just sticking to custom code.
So I uninstalled it. It was a tough call; going “backward” isn’t something founders like to do. Now the frontend is simpler, lighter, and more under my control.
Can you talk about building the article insights tool from scratch instead of using third-party libraries?
I don’t know if some of the existing libraries are even built by experts in the digital media or writing space, either way, they didn’t seem to be exhaustive to what I wanted to build – which was literally a replacement of sorts to actual editorial advice.
As a former editor and digital media founder, I spent years giving feedback to writers, to maximize views, engagement, improve clarity, SEO, tone, clickability. While building Toxt, I realized I would need to automate this editorial instinct for young writers starting out as I won’t be able to give them feedback while I’m busy coding! Instead of relying on generic sentiment libraries or black-box AI APIs, I built my own system using custom word lists and complex rule-based algorithms, all based on my real editorial experience.
Not just that, many Thoxt writers and our interns have given me feedback on how helpful or easy-to-understand these generated insights are, which has helped me improve it further. It’s not fully instinctive yet, but it’s getting there.
This way, the tool doesn’t just spit out vague scores, it offers practical, specific, and contextual insights for writers.
What does the insights tool analyze, and how does it help writers improve?
The tool gives plenty of feedback on readability, skimmability, emotional tone, bias, SEO alignment, AI content detection, and more. We also analyze title strength and offer engagement prediction based on historical patterns from my media experience.
The goal isn’t to “grade” writing, but to get writers to think, sort of when they would discuss their writing with me. Whether it’s spotting emotional bias, predicting engagement, or flagging a weak title, the system mirrors how I’d actually review a draft. It turns Thoxt into more than a publishing tool, it becomes an upskilling platform that aims to make writers confident.
Whether someone’s a journalist or just publishing their first thot, Thoxt helps them learn, iterate, and grow, without needing a human editor hovering over their shoulder. Writing should be fun, and that’s what Thoxt is all about, but if writers aim to improve and evolve, Thoxt helps with that too.
What features were deceptively hard to build?
Drafts and image uploads. I always knew the editor would be the most important and toughest part of Thoxt to build and so I started coding Thoxt with it. Everything’s fun and dandy with text and simple formatting and an image or title. But all hell breaks loose when you want to save settings like roles, detect their city and timezone, add video optimizations and generate+save 1-click meta-data/SEO-automation and much more all with a single saving point, and auto-save on top of that without re-loading the page once so it’s all seamless while users are trying to flow their thoughts on the page!
Syncing image uploads with unsaved content is tricky, so is saving different settings for editors and writers, while managing scheduling in different time zones and notifications. Building that logic without race conditions took way more planning than expected.
How do you handle drafts, autosaving, and image uploads without race conditions?
Well, you have to ensure the article object exists before anything else, even before the user starts typing, so there’s no unnecessary page reload upon the first save; but at the same time avoid filling a user’s dashboard with unintended ghost drafts. Autosave happens in the background as the user writes, and image uploads are linked to that draft ID. This lets images attach in real time without waiting for a manual save; all while remembering their order within the draft, even if the user moves images up or down it as the article evolves. It’s invisible to the user but super intentional in the backend.
So many times it got more complex in the back-end due to making things simpler for the user on the front-end, but that ease of publishing has always been the top priority.
What was your biggest “aha” moment during the build?
It kind of became clear during the software design phase itself. That a futuristic writing tool can’t feel like software. Users treat the editor more like a canvas for ideas, so it needs to cater to a variety of use cases – adding sources, GIFs, code blocks, embedding videos and links, some will write short, some will write longer articles. Some will prefer adding image galleries, polls, quizzes, videos or even doodles. Some will add just text and expect you to make it look aesthetic.
So while old tools felt more like “form with fields,” modern writing should feel like posting on social media – intuitive, easy and fast. You can’t expect them to “look for things” or take a tutorial for this – it just has to be obvious.
Post-AI Era Development
Have you used AI tools (like ChatGPT or GitHub Copilot) during development? How?
Yes – mostly just ChatGPT for second opinions on architecture and Copilot for help with debugging or writing code and, you know, auto-complete. ChatGPT has also been helpful with understanding new libraries or prototyping features, like image or video optimizations. They’re great for learning, acting as a second-brain. It replaces the need for a team at the early stages.
You’ve spoken out against vibe-coding. What do you think is wrong with it?
I just feel that vibe-coding overhypes the current capabilities of AI. We’re just not there yet. Even if you try to generate a feature or simple app with more than one basic feature via a single prompt (without even looking at the code), it feels like a waste of time going in circles trying to fix its bugs or getting it to work the way you envisioned it.
For early-stage founders, time is of essence and time should be invested in building code that can be evolved and improved upon. Everything needs to be well-thought of and optimized the best you can, so you can spend the next day building the next feature instead of debugging the previous ones. That’s how it remains fun.
So when I spend my time on planning architecture, thinking of algorithms and discussing with ChatGPT on how we should go about building a feature instead of just whipping it up and letting it be – it feels like I’m investing my time rather than wasting it.
Indie Dev/Founder Philosophy
How has AI changed the way solo founders or small teams can build platforms?
For the older tech giants, having a large workforce used to be a flex. Now I feel that having a small team will be trendy again. You can be nimble and iterate faster if you don’t have a bloated team. You don’t need even investment unless you need a team. Founders don’t need to waste time on hiring and firing and finding VCs anymore, they can just focus on building their product and talking to customers.
So bootstrapped 1-2 person teams can become a thing again, and that means better products, faster improvements. AI also means 1 person teams within the startups, as it increases productivity across domains, so becoming an early-stage startup employee would be a great idea now for those who love working alone or have a mind of their own.
What advice would you give someone building a full-stack product alone in 2025?
Start with the real problem you care about, get as much feedback from your potential users or customers as you can before and while you build. It’s easy to get overwhelmed by tech trends or feel like you need AI in everything, but the best tech companies have simply made the current solutions better, built what the founders themselves wanted to exist or use. Build what you understand, and improve as you go. Avoid naysayers on social media.
Also: spend more time on planning the back-end architecture, modularize early, and don’t be afraid to undo things that aren’t working. And finally, treat every small feature as a chance to make the product feel joyful. That’s what keeps you going when it’s just you, your code, and a very big vision.
Tech journalist uncovering innovations shaping the future of India. I’m your chap for tech entrepreneurs interviews, inside scoops, latest in coding and dev ops.