Malhar Inamdar
research, machine learning and artificial intelligence

⛰️ Among the hills of Mussoorie, Uttarakhand
Hey, I’m Malhar, a third-year undergrad at PICT, Pune, working on AI systems that can help solve complex problems and benefit humanity in meaningful ways.
My research interests center on impactful applications of AI such as across climate, healthcare, reasoning models, and multilingual language modelling. I’m particularly drawn to building systems that improve accessibility in low-resource settings and interpretability that improves model trust and reliability - whether that means developing diagnostic tools for underserved communities or creating language models, frameworks for mutlilingual settings.
What I’ve worked on
Vaidya Nidaan — Led the development of an Alzheimer’s diagnostic platform that integrates CNN-based analysis with medical imaging using FSL biomarker identification (hippocampal volume, white/gray matter ratios). The system employs GradCAM for visual interpretability and includes a multilingual RAG pipeline that generates structured medical reports grounded in research literature. This work secured third place among 400+ teams at PICT Techfiesta 2025.
Medical Reasoning with Phi-3 — Fine-tuned Phi-3-mini-4k-instruct on the MedQA-USMLE dataset to teach structured clinical reasoning. Using QLoRA, I trained the model to generate explainable diagnostic chains-of-thought while engaging only 0.44% of its parameters. To explore preference-based alignment further, I implemented a GRPO trainer from scratch in PyTorch, working through the practical challenges of low-precision optimization.
Regional TinyStories — At Vizuara AI Labs, I conducted research on multilingual language modeling under Dr. Raj Dandekar, extending Microsoft’s TinyStories (2023) work for Indian regional languages. I trained Small Language Models (2M–150M parameters) from scratch for Hindi, Marathi, and Bengali. We developed a novel framework for the development and analysis of SLMs, tokenizer performance, linguistic complexity, machine translation performance and demonstrated that a 53M parameter model could achieve GPT-3.5-comparable results on short-story generation - a promising direction for accessible, high-quality models in Indian languages. This work is currently under review at ACL ARR 2025.
At Froncort.AI, I implemented a RLHF pipeline that converted expert reviewer feedback into heuristic reward signals, improving LLM output quality by 35% while working within tight compute constraints compliant with international regulatory and ontological standards; and architected multi-agentic system for regulatory document generation workflows.
When I’m Not Training Models..
I’m probably reading papers, books at odd hours or wandering through the hills. I’m drawn to the philosophical questions that emerge from AI’s capabilities, but I’m most energized by the opportunity to build systems that genuinely help people. Music often accompanies this process — it’s become a quiet companion that helps me think through problems and find clarity.
If you’re working on something in healthcare, climate AI or language modeling or just ml in general, feel free to reach out! I’d be glad to hear from you.
news
Jul 24, 2025 | Selected for the highly competitive Data Science: Probabilistic and Optimization Methods II program of International Centre for Theoretical Sciences - Tata Institute of Fundamental Research featuring lectures by leading experts on data science, probabilistic models, and optimization techniques. |
---|---|
Jun 14, 2025 | Was selected to attend Microsoft Research India's Academic Summit 2025 virtually, a highly selective research summit held for researchers, academic professors and PhD, Master's and undergraduate students across India. |
Apr 14, 2025 | Regional Tiny Stories: Using Small Models to Compare Language Learning and Tokenizer Performance — preprint available on arXiv. |
Feb 11, 2025 | Our project focusing on the health of Alzheimer patients Vaidya Nidaan stood 3rd in Pune Institute of Computer Technology (PICT) Techfiesta Hackathon out of more than 400+ teams. Project Link |
Apr 21, 2024 | Stood 2nd (Runner-up) in Cretronix, a multi-stage contest involving electronic circuit design and microcontroller programming, conducted as part of Credenz 2024, the annual technical fest of PICT IEEE Student Branch. |