Data science and machine learning. M.Tech from IIT Guwahati, where my thesis classified 15 hand postures from 8-channel surface EMG.
The thing I took from that work: the evaluation protocol decides the number. Correcting a leaky data split in my own thesis cost 14.5 accuracy points, and the corrected figure is the one I report. My projects spend as much effort on how a result was validated as on producing it.
Toolkit — Python (pandas, NumPy, SciPy, scikit-learn) · SQL / PostgreSQL · signal processing · embeddings and vector search · Tableau · Jupyter
| Project | What it does | Stack |
|---|---|---|
| sEMG Hand Posture Classification | M.Tech thesis — 15 postures from 8-channel EMG. 83.7% within-subject and 32.3% cross-subject against a 6.7% chance baseline; a leaky split had inflated the headline to 98% | Python, scikit-learn, NMF, signal processing |
| Semantic Book Recommender | Natural-language book search over a vector database, with zero-shot category and emotion filtering, served through a Gradio app | LLMs, embeddings, ChromaDB, Gradio |
| Plant Disease Classification | CNN image classifier that detects plant disease from leaf photographs | Deep learning, CNNs |
| Customer Segmentation — RFM + K-Means | Segments online retail customers into high-value and at-risk groups to target marketing spend | Python, scikit-learn, pandas |
| Sales & Customer Analytics with SQL | 15+ business questions answered on a digital music store database — customer behaviour, sales trends, revenue and genre popularity | PostgreSQL, CTEs, window functions |
| Hospital Mortality Prediction | Analyses patient records to estimate in-hospital mortality risk | SQL, Excel |
Email — harshitp0998@gmail.com
LinkedIn — linkedin.com/in/harshitparamhansofficial