Manav
Bhuta
Building intelligent systems that matter.
Computer Engineering student at NMIMS, Mumbai — with 3 published ML/AI papers across IEEE WCSC, DACS, and OTCON. I research LLM evaluation, reinforcement learning, and predictive modeling.
The Story So Far
I'm a 3rd-year Computer Engineering student at NMIMS, Mumbai, working at the intersection of machine learning research and engineering. I don't just study AI — I publish about it.
My research spans LLM evaluation in RAG systems, carbon-aware reinforcement learning, and predictive ML pipelines. I've had work accepted at IEEE WCSC 2026, DACS 2025, and OTCON 5.0.
Outside research, I've led ML events as Sub-Head at DataMavericks, mentored peers, and competed as a Datathon finalist at Analytika NMIMS.
I'm actively seeking ML/AI internships where I can deploy intelligent systems at scale — not just build them in notebooks.
Published Work
Comparative Evaluation of LLMs in Multi-Document RAG Systems
Developed a rigorous evaluation framework comparing GPT-3.5, Claude-3-Haiku, and Gemini-Pro within a retrieval-augmented generation pipeline over a 172-page enterprise corpus. Achieved statistical significance (p < 0.05, Cohen's d > 0.96). Key finding: Claude exhibited highest faithfulness (0.904) with 40% honest refusals, while GPT and Gemini prioritized 100% coverage.
Box Office Revenue Prediction using Machine Learning
End-to-end ML pipeline for pre-release box office revenue forecasting using real-world TMDB data. Benchmarked 3 regression and 3 ensemble models with feature engineering and hyperparameter tuning. XGBoost achieved best performance: R²=0.77, MAE=$43.9M.
Eco-Scheduler: A Carbon-Aware Intelligent Workload Distribution Framework
Reinforcement learning-based decision system for workload allocation across multi-cloud environments. Framed as a Markov Decision Process optimizing cost, latency, and resource utilization trade-offs. Enables carbon-aware compute orchestration for sustainable AI infrastructure.
Built, Shipped, Learned
Multi-Document RAG System
A retrieval-augmented generation system for multi-document QA with semantic chunking, dense retrieval, and a custom evaluation framework. Published at IEEE WCSC 2026.
Engineered novel evaluation metrics achieving statistical significance across three frontier LLMs. Key insight: measurable coverage-vs-faithfulness tradeoff in RAG — Claude prioritizes reliability, GPT/Gemini prioritize completeness.
EmoSound: Mood-Based Music Recommender
Real-time emotion detection from voice with music recommendation. 87% accuracy across 7 emotion classes using Librosa + Spotify API.
EcoFinanceAI
NLP pipeline linking financial investment data with environmental impact reports via transformer-based entity linking. Built during Research Internship at Suvidha Foundation.
Box Office Revenue Predictor
End-to-end ML pipeline for pre-release revenue forecasting. XGBoost won with R²=0.77 and MAE=$43.9M. Published at DACS 2025.
Roles & Achievements
Let's build something
important together.
Actively looking for ML/AI internship opportunities. If you're working on something interesting in AI research, intelligent systems, or applied ML — I'd love to talk.