Dev Bansal

IN · trying stuff :p
ml · nlp · cv · agents

about

19. second year undergrad. i read papers and build things. that's it.

projects

Sharpner — no-code AutoML platform

build, train, and compare ml models through a ui. no code required.

kenkyū — autonomous AI research agent

plans queries, searches the web, and synthesizes reports using langgraph + ollama.

anzen 安全 — real-time hazard detection

detects vehicles, pedestrians, and road hazards with severity-based alerts via yolov8.

paper-inspired builds

ResNet18 transfer learning on CIFAR-10

fine-tuned resnet18 on cifar-10 using imagenet weights after reading the resnet paper.

anzen 安全 — real-time hazard detection (YOLOv8)

applied yolov8 to road safety after working through the original yolo paper.

paper implementations

AlexNet — Krizhevsky et al., 2012

from-scratch pytorch implementation of the architecture that started the deep learning era.

ResNet — He et al., 2015

implemented residual connections from scratch; trained on standard benchmarks.

YOLOv1 — Redmon et al., 2015

full from-scratch implementation of single-shot object detection in pytorch.

writing

building the brain logic of ai agents: a beginner's guide

a breakdown of how ai agent logic works and how to build it — reposted by the founder of langchain.