Machine Learning Engineer

PORTFOLIO · MACHINE LEARNING ENGINEERING
Machine Learning
Engineer
Building deployable AI systems that connect models, APIs, products, and measurable outcomes.
Bangalore, India · Python · FastAPI · TensorFlow · XGBoost · AWS · Docker
PROJECT 01 · REAL-TIME TRAFFIC ANALYSIS
Traffic intelligence starts with live data
View project folder →
Problem: Surface congestion in Bangalore from live traffic conditions.
Approach: TomTom flow data → feature engineering → model comparison → FastAPI + Streamlit + Folium.
Tools: Python, Pandas, scikit-learn, XGBoost, SVM, Plotly, Docker, GitHub Actions.
Finding: The best model is persisted for dashboard use; the repository does not report an accuracy metric.
PROJECT 02 · AWESOME STUDENT RESOURCES
Make student opportunity easier to find
Free tools, courses, software, and student perks.
11 practical categories, from cloud services to career tools.
22 stars and 11 forks signal community pull.
View project folder →
Confidence-aware vision for apple varieties
96.22%
Validation accuracy
98.1% of predictions cleared the confidence gate across 10 market varieties.
Model & decision gate
EfficientNetV2S with two-stage transfer learning; predictions below 75% confidence return “ambiguous.” Tools: TensorFlow/Keras, Flask, Streamlit, Docker, Vercel, Jupyter.
View production app folder →
PROJECT 04 · OBJECT DETECTION
Turn video into trackable movement



YOLOv8n detects 80 COCO classes. ByteTrack assigns persistent IDs to preserve identity and motion context across frames; the app returns annotated video, trajectories, and heatmaps.
Sample: 40 frames · 133 detections · 15 unique track IDs; processed in 0.57 seconds.
Tools: PyTorch, OpenCV, Flask, Gunicorn, Bootstrap, pytest.
View project folder →
PROJECT 05 · ACTIVE RESEARCH
Researching smarter video captioning
Workflow
Reduce video-captioning cost: select fewer informative frames, ranked by motion, semantic novelty, and temporal diversity. BLIP-2 captions selected keyframes; fuse captions into a video summary.
Evaluation design
Compare six sampling strategies on MSVD at a fixed frame budget, with CIDEr and METEOR in the scoring plan.
Status: Active research direction; no results are committed yet.
View project folder →
PROJECT 06 · TRAFFICSENSE AI
From traffic prediction to urban decision support
One resilient platform connects congestion, weather, routing, and short-horizon forecasts for Bangalore.
25 corridors
Forecast speed 15, 30, and 60 minutes ahead.
R² 0.8911
MAE 4.578 km/h · RMSE 5.986 km/h.
77 tests passing
Deterministic demo data keeps the product resilient when APIs lapse.
View project folder →
The engineering toolkit behind the models
Modeling
scikit-learn · XGBoost · TensorFlow / Keras · PyTorch · EfficientNetV2S · YOLOv8 · ByteTrack · BLIP-2 · DINOv2
Productization
FastAPI · Flask · Streamlit · Next.js · React · Leaflet · Recharts
Delivery
Docker · Docker Compose · Vercel · AWS Bedrock · GitHub Actions · pytest
THE THROUGH-LINE
A portfolio built around usable AI
Live data systems
Move from ingestion to an interface people can use.
Uncertainty-aware vision
Treat confidence as a product decision, not a hidden detail.
Explicit research
Make sampling strategies and evaluation plans visible.
Production discipline
Deployment, tests, and fallback paths are part of the ML work.
Best fit: Machine Learning Engineer roles spanning applied ML, computer vision, and production AI systems.
NEXT SYSTEM
Let’s build
the next system
GitHub
github.com/Shashwat-19
LinkedIn
linkedin.com/in/shashwatk1956
Email
[email protected]