Electrical engineering × machine learning
Abhishek
Singh Rajawat
I build systems that turn real-world signals into accountable intelligence.
Abhishek
01 / The grid remembers
7.5 years
of demand history.
Weather, calendar and grid signals — joined into one living system.
02 / Prediction
3.56%
MAPE.
Across 203 chronologically held-out days. Forty-six percent less error than the same-weekday baseline.
03 / Proof loop
Forecast first.
Grade it publicly.
Every prediction is logged before reality arrives — then the model scores itself in public.
04 / Convergence
From current
to intelligence.
Forecasting · computer vision · NLP · automation
Explore the workProfile 2026
I build models that
answer to reality.
Electrical engineering taught me to respect physical systems. Machine learning taught me how to find their hidden patterns.
I’m a third-year B.Tech student at Jabalpur Engineering College, focused on forecasting, deep learning and automation that survives outside a notebook.
Selected system 01 / 05
Live Electricity
Demand Forecaster
An unattended forecasting system for Madhya Pradesh that collects, predicts, publishes and publicly grades itself every day.
3.56
% MAPE
-
01
Acquire
Grid-India reports and city-level Open-Meteo weather arrive on a scheduled pipeline.
-
02
Structure
Seven and a half years of demand, weather, holidays and cooling-degree signals live in PostgreSQL.
-
03
Forecast
LightGBM learns lag, rolling-mean and calendar structure across chronologically held-out days.
-
04
Prove
Predictions are committed before the outcome and graded with a public rolling 30-day MAPE.
46%error reduction vs. baseline
203held-out evaluation days
50pytest checks after API migration
Project field 02–05
Different signals.
The same discipline.
02
Natural language processing
Smart Review
Analyzer
Five-star rating prediction across 568K+ Amazon reviews using TF-IDF bigrams and four compared classifiers.
- Dataset
- 568K+
- Weighted F1
- 0.65+
- Classes
- 5
03
Computer vision
Fashion
MNIST CNN
A regularized convolutional network trained across 70,000 images with augmentation, BatchNorm and dropout.
- Test accuracy
- 91%+
- Classes
- 10
04
NLP classification
Email Spam
Detection
Precision-aware email classification with TF-IDF, stemming and three benchmarked model families.
- Accuracy
- 95%+
- Emails
- 5.5K+
05
API automation
YouTube
Analytical System
OAuth 2.0 and three Google APIs turn raw channel data into TF-IDF insights, generated Slides and delivered email reports — without manual intervention.
- Google APIs
- 3
- Manual steps
- 0
99.43%Andrew Ng supervised ML
14professional certifications
27GitHub repositories
5+end-to-end ML systems
Technical practice Toolkit
Built across
the full loop.
Languages
Python · C++ · SQL
Machine learning
TensorFlow · Keras · Scikit-learn · LightGBM · CNNs · time-series forecasting
Data systems
PostgreSQL · Pandas · NumPy · feature engineering · EDA · TF-IDF
Shipping
Streamlit · GitHub Actions · scheduled pipelines · OAuth 2.0 · Google APIs
Formation Education & recognition
Learning with
evidence attached.
2024 — 2028
Jabalpur Engineering College
B.Tech · Electrical Engineering · Third year99.43%
Supervised Machine Learning
Stanford / DeepLearning.AI · Prof. Andrew Ng99%
Databases for Developers
Oracle Dev Gym · Certificate of ExcellenceSelected
Smart India Hackathon
College-level selection round representativeOpen circuit Available for ML opportunities