Available for new projects

Hi, I'm Utsab.
I engineer & research the future.

I'm an electronics & information engineer specializing in computer vision, deep learning, and embedded systems. I build instruments that think — from U-Net segmentation to magnetic-resonance power transfer.

View my work arrow_forward Read research
~ $ whoami
utsab dahal -- engineer & ml researcher
~ $ skills --list
✔ Computer Vision / Deep Learning
✔ Python / PyTorch / TensorFlow
✔ Embedded Systems / ESP32
✔ Signal Processing & PCB Design
~ $

Areas of expertise

Computer Vision
Deep Learning
Embedded / IoT
Signal & Power
PCB Design
Research
0 Selected works
0 Peer-reviewed paper
0 Nepali HTR mAP@50
∞ Iterations in pursuit of clarity
I.

A brief, honest portrait.

Narrative / 01

I am a BEI passout from IOE Thapathali Campus, building a practice around the idea that rigorous engineering can feel editorial. My work ranges from U-Net brain-tumor segmentation to magnetic-resonance wireless power transfer.

I am an engineer who identifies the need, solves the problem and experiments until it works. I have a genuine love for hardware — oscilloscopes, coils, PCBs — and I am deeply interested in research and mathematics. If it can be measured, modelled or optimised, I want to be the one doing it.

  • Computer Vision
  • Deep Learning
  • Embedded / IoT
  • Signal & Power
  • Full-stack Tooling
II.

TechStack.

Stack / 02
code

Languages

Python · C · C++ · Java · SwiftUI · UIKit · HTML

neurology

ML & AI

CNNs · U-Net · MLPs · Segmentation · Classification · CV pipelines · Model evaluation

stacks

Frameworks

PyTorch · TensorFlow · Keras · OpenCV · NumPy · Pandas · Scikit-learn · Django · Next.js · React · MQTT · WebSockets

database

Data

MySQL · PostgreSQL · MongoDB · Firebase · Supabase

memory

Hardware & Sim

ESP32 · ESP-NOW · LTSpice · Ansys Maxwell · Custom PCB design

architecture

Foundations

DSA · OOP · OS · Networks · DBMS · System design

III.

The circuit bench.

Interactive / 03

I drew this rig on paper — V1, SW1, 470 Ω, 100 mH, 47 µF, D1 — photographed it, and taught the site to read it. Press test my work: Otsu thresholding, component detection, wire tracing, a clean vector redraw — and the sketch finally deploys as a live instrument you can drive.

Hand-drawn replica of the RLC bench circuit, ballpoint on paper
  1. 01 · raw scan
  2. 02 · otsu threshold
  3. 03 · detect & trace
  4. 04 · vectorize
  5. 05 · morph → bench
IV.

Selected works.

Archive / 2023 — 2026
01

Circuitron — hand-drawn circuit to live simulation

Turns a circuit photograph into a CircuitJS1 simulation using YOLOv7 detection at 95.6% mAP@50, handwritten-value OCR, skeletonized wire tracing and graph-based netlist synthesis.

YOLOv7OCRGraph ExtractionCircuitJS1
arrow_outward 02

MRC-RAG — grounded scientific research assistant

A citation-enforcing RAG system over 24 wireless-power papers, combining BGE embeddings, FAISS, BM25, reciprocal-rank fusion and cross-encoder reranking with abstention when evidence is insufficient.

RAGHybrid RetrievalRerankingFastAPI
arrow_outward 03

Nepali handwritten text recognition pipeline

End-to-end Devanagari handwriting recognition using YOLOv11 word detection at 96.6% mAP@50 and a CRNN-CTC recognizer with 96% character accuracy across 100 characters and 85K+ samples.

DevanagariYOLOv11CRNN + CTCPyTorch
arrow_outward 04

Fine-tuned TrOCR — handwritten component values

Fine-tuned a ViT-Transformer OCR model for handwritten circuit values, reaching 84.5% accuracy and 0.12 CER, then deployed batch inference and confidence scoring through FastAPI.

TrOCRVision TransformerOCRFastAPI
arrow_outward
05

Person Re-Identification — seven-camera CCTV forensics

A seven-camera CCTV system using YOLOv8 for person detection, ByteTrack and BoTTrack for tracking, CLIP for text-prompt-to-person retrieval, and VADCLIP for crime and anomaly detection across feeds.

YOLOv8ByteTrack + BoTTrackCLIP RetrievalVADCLIP7 CCTV Feeds
arrow_outward
Demo recording · 00:54 Seven-camera tracking, CLIP prompt search, ranked cross-camera matches and VADCLIP event detection
H.264 · 1600 × 872
06

True Wireless Charging — magnetic resonance

A mid-range magnetic-resonance power-transfer system with custom resonant coils and LC networks, empirically validated, peer-reviewed and presented at OKRP 2025.

Magnetic ResonanceCoil DesignPublication
arrow_outward 07

Brain-tumor segmentation with U-Net

A U-Net deep-learning pipeline for segmenting tumor regions from MRI scans, evaluated with Dice and IoU metrics.

U-NetMedical ImagingSegmentation
arrow_outward 08

IMDb sentiment classification with LSTM

An end-to-end text-classification pipeline with preprocessing and an LSTM sentiment model achieving more than 83% test accuracy on IMDb reviews.

NLPLSTMSentiment Analysis
arrow_outward 09

ThopaSichai — solar-powered smart irrigation

A solar IoT drip-irrigation system with ESP-NOW sensor nodes, a FastAPI backend and a Flutter control application.

IoTESP-NOWFastAPIFlutter
arrow_outward 10

ESP32 real-time monitoring system

An ESP32 sensor platform streaming live telemetry to a FastAPI backend over MQTT and WebSockets.

ESP32MQTTWebSocketsFastAPI
arrow_outward
V.

Publications.

Research / 04
KJSE · Vol. 10 May 5, 2026

Design and empirical validation of a mid-range wireless power transfer system using magnetic resonance coupling

Published in the KEC Journal of Science and Engineering (KJSE), Volume 10. Presented at the OKRP Conference 2025 on December 13, 2025.
DOI: 10.3126/kjse.v10i1.93856

Bench note 04 Physical prototype tested

Coil separation and lamp power

Slide the receiver coil along the ruled bench. This drawing explains the behaviour of the implemented apparatus; the paper contains the measured results.

d · Coil gap 70 mm
k/k₀ · Coupling 0.74
P/P₀ · Load power 54.7%
Observed state Coupled
Interactive resonant wireless power transfer model A supply drives a transmitter coil. Drag the receiver coil horizontally; the connected bulb dims as separation increases. AC source resonant drive L₁ — transmitter L₂ — receiver + lamp slide L₂ along the bench →
Reference marks
Fig. 1 Explanatory model; measured observations remain in the paper. B(z) ∝ (R² + z²)−3/2 · Prel ∝ B²
VI.

Education.

Provenance / 05
Mar 2022 — May 2026

IOE Thapathali Campus — Kathmandu

B.E. in Electronics, Communication & Information Engineering

Computer Programming · OOP · Probability & Statistics · DSA · DBMS · OS · Computer Networks · Calculus · Linear Algebra · Discrete Mathematics · Machine Learning

Open inbox

Initiate
dialogue.

Available for select freelance builds, research collaborations and full-time roles. I reply within a day, usually faster.