Final-year AI & Data Science student at K. J. Somaiya, Mumbai. I like shipping real products, testing them honestly, and writing down what I learn along the way.
I'm a final-year AI & Data Science student at K. J. Somaiya, Mumbai, and I care more about finishing things than starting them: a notebook that only works for me doesn't count. Two recent internships: at Makeflow India I built the session and conversation-flow logic for a WhatsApp-based AI wellness counselor; at Suvidha Foundation I fine-tuned six summarization models and iterated a PRIMERA pipeline's ROUGE scores upward. My final-year major project, PRISM, routes LLM queries between different hallucination-mitigation strategies and is currently under double-blind review at icSoftComp 2026. I led a six-member team at Smart India Hackathon 2025 and placed in the top 4% at the ZS Campus Beats Hackathon, and I compete on Codeforces and LeetCode when I have time. Right now I'm working on LLM evaluation and retrieval-augmented generation, and starting to learn model quantization and efficient inference. Open to internships and research collaborations.
Three rules I hold myself to
Test it honestly
If I can't say how something fails, it isn't done. Every project ships with the numbers that prove it works, and the ones that show where it doesn't.
Finish the thing
A prototype on my laptop isn't a product. I take ideas all the way to working apps, dashboards, and live pipelines that someone else can actually use.
Show the work
Sources over confidence. Causes over alerts. I'd rather explain exactly why something behaves the way it does than hide behind a black box.
- Sept 2025 - Nov 2025
Backend Developer Intern
Makeflow India · Nahur, Mumbai
Backend intern on Blumy, a WhatsApp-based AI wellness counselor.
- Built the core session management and conversation flow for Blumy, a WhatsApp-based AI wellness counselor: session initialization, stage-based routing (Concern, Solution, Feedback), and the introduction-to-concern transition, in Node.js.
- Built the introduction pipeline for new and returning users: conditional profile capture, returning-user shortcut paths, and keyword-based crisis detection that skips onboarding for urgent messages.
- Refactored the backend: split AI response generation into its own service, consolidated duplicate utilities, and restructured the conversation entry function so profile building no longer sat inside stage-specific session handling.
- May 2025 - Jul 2025
Machine Learning Intern
Suvidha Foundation · Nagpur, Maharashtra
Machine learning intern fine-tuning summarization models on real-world news datasets.
- Fine-tuned 6 summarization models (BART, Pegasus, DeepSeek-v1, FLAN-T5, T5, BigBird-Pegasus); BART performed best at ROUGE-1 0.44, ROUGE-2 0.20, ROUGE-L 0.30.
- Iteratively fine-tuned a PRIMERA multi-document summarization pipeline, raising ROUGE-1 from 0.36 to 0.44 and ROUGE-L from 0.18 to 0.20.
- Coordinated fundraising outreach that raised ₹20,000+ and supported 1,500+ community members across foundation campaigns.
- 2023 - 2027
B.Tech, Artificial Intelligence & Data Science
K. J. Somaiya School of Engineering · Mumbai, India
Final year. Led a 6-member team at Smart India Hackathon 2025. Top 100 of 2,450+ teams (top 4%) at the ZS Campus Beats Hackathon.
Machine Learning Specialization
Stanford University / Coursera (Andrew Ng)
PyTorch for Deep Learning
Udemy
Deep Learning A-Z: Neural Networks, AI & ChatGPT
SuperDataScience (Udemy)
AI Engineer Core Track: LLM Engineering, RAG, QLoRA, Agents
Ligency (Udemy)
DevOps: Complete Kubernetes Course
Udemy
The rest is better in person.
Or at least over email. Ask me anything: code, coursework, or coffee in Mumbai.