SUNANDA MANNE
B.E. AIML @ VTU · CGPA 9.1 · DRDO Intern · Top 5% of Department
Building intelligent systems at the intersection of research and production.
I'm a final-year AI & Machine Learning engineer with hands-on research at DRDO, where I fine-tuned multilingual transformer models and shipped 3× compressed inference pipelines. I love the full arc — from raw data and ablation studies to production APIs and dashboards stakeholders actually use.
Currently focused on NLP, model compression, and full-stack AI applications that solve real problems in healthcare, education, and finance.
{
"name": "Sunanda Manne",
"role": "AI/ML Engineer",
"university": "VTU",
"focus_areas": [
"NLP & Transformers",
"Model Compression",
"Full-Stack AI"
],
"currently": "Shipping research → prod",
"status": "open_to_work"
}Where I've shipped.
DRDO — Defence Research & Development Organisation
- — Fine-tuned mBART & mBERT on 2M+ sentence multilingual corpus (4 Indian languages, 5 domains) using PyTorch & Hugging Face — BLEU score 0.52 → 0.68 (+32%).
- — INT8 quantization + structured pruning compressed model 3× (1.8GB → 600MB), inference latency cut 40% (320ms → 190ms) — zero accuracy regression.
- — Built end-to-end pipeline processing 500K+ records/day; tracked 60+ experiments across BLEU, latency & robustness metrics.
Insight Driven Research
- — Applied Regression & K-Means clustering (k=5, Silhouette: 0.71) on 1,200+ survey responses to surface 3 high-signal adoption segments.
- — Designed 8 Tableau dashboards cutting stakeholder reporting cycle 60% (5 days → 2 days).
Bluestock Fintech
- — Built Python ETL pipelines (Pandas + SQLAlchemy) processing 50,000+ daily financial records; automated schema validation cutting data errors 35%.
- — Delivered 5 Power BI dashboards adopted as primary reporting tool by 12-person finance team.
Shipping intelligence.
Medi Predict AI
Multi-disease healthcare prediction platform — Heart, Diabetes, Kidney & Liver disease screening with real-time clinical recommendations.
- — Precision: 0.89 · F1: 0.88 across 4 disease models
- — 50+ automated healthcare recommendation workflows
- — JWT-secured patient sessions with PDF report generation
AI-ML Attendance System
Facial recognition at scale across 40 classrooms — sub-2s identification, <1.8% false-positive rate.
More on GitHub
Research notebooks, side projects, and ablations live in my repositories.
Explore Repos →The toolkit.
Credentials.
OCI 2025 AI Foundations Associate
Python for Data Science
Forward Program
Data Analytics (4 courses)
Data Fundamentals
Let's Build Something Intelligent.
Whether it's a research collab, full-time role, or just a question about transformers at 2am — my inbox is open.