Chitranshu
Harbola.
I build things that shouldn't exist yet — AI systems, learning tools, and small bets that grow into companies. From a hill town in Uttarakhand to a conference stage in Lecco, Italy, by way of IIT Madras.
I grew up in Ranikhet, a small town in the Kumaon hills. No one around me was building software companies. So I started one of my own and figured it out from there.
In Class 9, I ran a gaming YouTube channel to 20,000 subscribers — then burned it down because it felt unserious. I taught myself Python, started teaching data science to engineers twice my age on Fiverr, broke into my school's website at 15 to train a grade-prediction model (I don't recommend this), and have been chasing some version of that same impulse ever since.
Today I'm a Founder & CTO, I write papers with my professor at IIT Madras, I've shipped AI systems for enterprises and universities, and I just flew to Italy to present research. I'm 21.
I started by trying to predict grades in Class 10. Today I'm building AI systems, publishing research, and running startups. I'm just getting started.
Gaming YouTuber → self-taught programmer Pivot #1
Hit 20K subs on gaming, got bored of it, pivoted the channel to Python tutorials. Views collapsed. I didn't care — it was the first time building felt like it mattered.
My first ML model (and a hard lesson) Origin
Found a vulnerability in my school's site, pulled 800 students' report cards, built a grade-prediction model with my CS teacher. Learned ML. Also learned something more important about consent and data.
Teaching data science to people twice my age
Picked up clients on Fiverr. Half of them were working engineers with 10+ years of experience. Teaching them forced me to actually understand what I was saying. My pricing arc: $10/hour → $100 for 10 sessions → a $150 course, which I eventually published publicly on YouTube.
Freelancing across four continents
Shipped websites, booking systems, retail backends, and a UK roster-management app where I pitched the client an LLM-based player-evaluation layer they hadn't asked for. They took it. That's when I knew I liked being more than an implementer.
MiniZone — an Amazon for a hill town Startup #1
Built a local marketplace with a Ranikhet shopkeeper. Learned the hard thing: a working app isn't a working company. Distribution, trust, and behavior change are the actual product.
Fine-tuning SLMs for NOMI · CultureVo US client
Worked with Ajita Agarwala (ex-IAS), co-founder of CultureVo in the US. Fine-tuned small language models for NOMI, their personal-companion product. Scraped corpora, trained, iterated. Learned the deep lesson: datasets, not model size, decide most outcomes. ₹10K/month.
Vidyarang — AI teaching chatbot Team of 8
Built Vidyarang with Anupam Purwar (ex-Senior Scientist, Amazon; ex-Director of ML, Sprinklr). I led a team of eight — 2 UI/UX designers, 1 backend dev, 1 ML engineer, and 4 interns from BITS Pilani. We wrote three papers together, presented at the ISB CODE conference in Varanasi (January 2025) and at I4AM in Bangalore (Springer). Then NotebookLM shipped and quietly ate the problem we were solving. We stopped. A clean lesson in moats and timing.
Applied ML with Debdoot Mukherjee Industry
Worked directly with Debdoot Mukherjee (Head of Demand Engineering, Meesho) on an applied ML project — pipelines, API extraction, cleaning, transforms, validation. The unglamorous plumbing that makes everything else possible, learned from someone who had seen it built at Meesho scale. ₹25K/month.
Full-stack LLM app · University of Warwick, UK via Aigurukul
Met Prof. Shweta (University of Warwick, United Kingdom) at the ISB CODE conference in Varanasi in January 2025. By April we were building together; shipped through September. An avatar-driven survey-analysis platform — full-stack: AI, humans, data, and shipping, all at once. Paid through Aigurukul: a MacBook Air and ₹60K.
Led sales automation for DesiCrew / QAonCloud ₹1.12 L · solo build
Walked into DesiCrew's office at the IITM Research Park, scoped the automation their sales team needed, wrote the proposal, and led the build end-to-end. Over four months I shipped an AI SDR system — company discovery, Apollo enrichment, LLM scoring, LinkedIn outreach, FastAPI + Supabase + Gemini. Their professional QA engineer surfaced 150+ UI issues; I fixed every one, alone. ₹1.12 lakh for the engagement.
SpiritAI — my first DeepTech company Startup #2
Founder & CTO. Conversational AI around spiritual experiences. Incubated at IITM Nirmaan. Learned that technical capability ≠ product-market fit — and that being wrong early is cheaper than being wrong late.
Gamifai — the pivot Pivot #2
Turned toward edtech: convert textbook problems into interactive, gamified levels. The lesson from SpiritAI stuck — be loyal to the problem, not the idea.
IIT Madras DSAI Lab · Prof. Balaji Srinivasan ₹10K/mo · IEEE · patent
Applied LLM research — knowledge graphs, RAG, learning systems, industrial AI. With Prof. Balaji, published 1 IEEE paper (presented at AIMLSystems 2026 in Lecco, Italy) and filed a patent. TA for the Speech & Language Technology course. ₹10K/month stipend. Travel to Italy sponsored by my professor; stay by Politecnico di Milano and the COMSNETS association; conference registration by IITM Nirmaan. Nirmaan also funded my I4AM presentation at IISc Bangalore.
From Papers to Animations: A Plan-Mediated LLM Pipeline with Parser-Friendly Code Generation
An LLM pipeline that turns research papers into educational animations — plan first, then generate parseable animation code. Presented in Italy.
Politecnico di Milano · 2026LLM-Aided Machine Prognosis
Can LLMs reason over long sequences of numerical sensor data? We serialized frequency signals from industrial bearings and tested anomaly detection. Yes — if you represent them right.
Industry 4.0 & Advanced Manufacturing · 2026Prescriptive Agents based on RAG for Automated Maintenance
LAMP told you something broke. PARAM tells you what to do about it — a RAG-plus-agents framework for prescriptive industrial maintenance.
Poster · 2025Knowledge-Oriented Small Language Models
Exploring how smaller language models can carry structured knowledge in domains where general-purpose LLMs are overkill.
2026Build before you're ready. Curiosity ships faster than expertise.
Distribution is a skill. 20K subs taught me that nobody discovers a good product by accident.
Fall in love with the problem. Not the first idea you had about it.
Research and engineering share a bloodstream. I like living in both.
Teaching compounds. You can't explain what you don't understand.
Geography is not a ceiling. Ranikhet to Milan is a straight line if you keep walking.
AI-native products. Agents that actually do work. Learning systems that teach the way good teachers do. Developer tools for people solving hard problems. I enjoy the whole stack — the research question, the model, the pipeline, the backend, the UI, and the person who finally hits deploy. I don't want to be only a researcher, only an engineer, or only a founder. I want to keep being all three.
A second newsroom. Short, honest clips about building, research, and the arc from Ranikhet to Milan.
Let's talk.
I'm always up for a conversation about AI, research, startups, or the strange paths that led us there. If you're building something interesting — reach out.