VOL. 0125 connected parts
From neurons to intelligent machines
One connected book: begin with electrical intuition, build computational models, and learn how to test ideas about intelligent machines.
By Harshit MakwanaRead the book →
THE RESEARCH LAB
Follow the question.
Build to understand.
A map of technical interests and documented project work. Published case studies are the source of the explanations here; future research areas are clearly marked.
Explore the knowledge universe ↗THE AUTHOR’S SHELF
Books by Harshit.
Big ideas, explained patiently. A growing collection of illustrated books on the things I find worth understanding—one clear example at a time.
Computer Vision
Detection, tracking, segmentation and image restoration, grounded in the approaches documented in my project case studies.
Edge AI
Inference optimization and deployment constraints, including the OpenVINO experiments described in Roadway Solutions.
Robotics
Flight control and sensor management using ESP32, LoRa and MicroPython.
Physical AI
Explore connections to sensing, vehicle monitoring and embedded systems. Dedicated research notes have not been published.
Autonomous Systems
A space for future notes on perception and control. The linked drone project documents embedded flight-control work, not a claim of autonomous capability.
Agentic AI
A space for future research notes and experiments. No results have been published here yet.
Research notes will appear when published.
Digital Twins
A space for future work on models of physical systems. No dedicated experiment is published here yet.
Research notes will appear when published.
Reinforcement Learning
A space for future notes on learning from interaction. No research results are claimed here.
Research notes will appear when published.
Multimodal AI
A space for future notes connecting different kinds of input. No dedicated experiment is published here yet.
Research notes will appear when published.
Experiment directory
No standalone research experiments or new results have been published here yet. The technical notes below come from existing project case studies.
DOCUMENTED APPROACHES
Inside the engineering.
Technical notes drawn from existing case studies. These are project summaries, not new research publications.
Data engineering
Scalable Data Fusion Engine
I implemented data-fusion logic and ETL processing, then exposed data through FastAPI REST services. PostgreSQL and Trino supported storage and querying, Redis handled real-time streams and caching, and Solr provided search and indexing capabilities.
Read the technical approach ↗Computer vision
VehiSense Pro
I fine-tuned the detection model and worked on the temporal logic needed for vehicle tracking, speed calculation and U-turn detection. I extended the workflow with incident and parking features so the application could present more than isolated detections.
Read the technical approach ↗Computer vision
Roadway Solutions
I used fine-tuned YOLO for detection and segmentation and ByteTrack to connect detections over time for tracking and counting. For potholes, I combined the segmented object contour with MiDaS depth estimates to support analysis of the detected road surface.
Read the technical approach ↗Natural language processing
Automated Resume Insights
My implementation focused on turning resume text into organized fields. The NLP workflow identified key information and grouped it into categories that correspond to common recruitment review tasks.
Read the technical approach ↗Computer vision
Image Inpainting Using GANs
I used artificially created masks to simulate missing image regions in the training dataset. This gave the model examples centered on the reconstruction task rather than treating image generation as an unconstrained problem.
Read the technical approach ↗Geospatial analysis
GIS Conflict Detection and Prediction
I started with data preparation: bringing roadway line strings and trip coordinates into a workflow where I could clean records, inspect outliers and carry out exploratory analysis. Shapely and GeoPandas supported the spatial operations that preceded modelling.
Read the technical approach ↗Transport analytics
Turning Movement Count Estimation
I combined data acquisition with database-side preprocessing and Python mapping. This work brought road records and their associated values together before modelling, while seasonality analysis informed how I interpreted unusual observations.
Read the technical approach ↗Natural language processing
AI-Powered Interview Bot
I separated the conversation into a general aptitude and reasoning stage and a job-specific stage. This gave the interview flow an explicit structure rather than treating every question as part of one undifferentiated conversation.
Read the technical approach ↗Voice interfaces
Voice-Assisted Chatbot Integration
I implemented Siri interaction through App Intents in the Swift application. On Android, I combined a Flutter/Dart application with native Kotlin App Actions integration and Dialogflow ES. Each platform had its own implementation while serving the same travel-information use case.
Read the technical approach ↗