1) MediScript – AI-Powered Voice-to-Prescription System
Tech: MERN Stack, Python, NLP, Speech-to-Text, PDF Generation, REST APIs, Wishper, Transformers, Bi-LSTM, CRF
- Designed and developed MediScript, an AI-powered voice-to-prescription system for clinics and telemedicine workflows.
- Enabled doctors to generate structured medical prescriptions from voice input, reducing manual documentation effort.
- Implemented NLP pipelines to extract medications, dosage, frequency, and clinical instructions from speech transcripts.
- Integrated telemedicine functionality for online consultations, supporting remote patient care.
- Generated downloadable, standardized prescription PDFs with patient and doctor metadata.
- Focused on privacy-first design, multilingual support, and accessibility for Indian healthcare settings.
2) Retail & Marketing Analytics – Customer Segmentation & Growth Strategy
Tech: Python, SQL, Pandas, Scikit-learn, Power BI / Tableau
- Conducted customer segmentation analysis using transactional retail data to support targeted marketing strategies.
- Implemented RFM (Recency, Frequency, Monetary) analysis to quantify customer value and purchasing behavior.
- Applied K-Means and hierarchical clustering, validating clusters using silhouette scores.
- Translated analytical findings into actionable business strategies for customer retention, cross-selling, and churn reduction.
- Estimated revenue uplift scenarios based on improved conversion and retention rates for high-value segments.
3) Financial Operations Analytics – Revenue Forecasting & Churn Analysis
Tech: Python, Pandas, Prophet / ARIMA, SQL, Scikit-learn
- Built an end-to-end financial analytics pipeline using large-scale e-commerce transaction data (Olist).
- Analyzed monthly revenue trends and seasonality, forecasting future revenue using time-series models.
- Defined and modeled customer churn behavior based on purchase gaps, delivery delays, and review scores.
- Performed profitability analysis by incorporating logistics costs, delivery delays, and return rates.
- Conducted what-if analysis to evaluate business impact of reducing delivery delays and improving customer experience.
4)Fine-Tuning Open-Source Large Language Models
Tech: PyTorch, HuggingFace Transformers, PEFT (LoRA), Tokenizers
- Trained and fine-tuned transformer-based language models on domain-specific datasets using HuggingFace and PyTorch.
- Implemented parameter-efficient fine-tuning (LoRA) to reduce compute and memory requirements.
- Designed custom tokenization and data preprocessing pipelines for instruction-style datasets.
- Evaluated models using perplexity, BLEU, and qualitative generation analysis.
- Performed hyperparameter tuning to balance convergence speed, generalization, and inference efficiency.
