<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Projects on Niraj Thapaliya</title><link>https://ndthp.com/projects/</link><description>Recent content in Projects on Niraj Thapaliya</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Fri, 01 May 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://ndthp.com/projects/index.xml" rel="self" type="application/rss+xml"/><item><title>Context-Aware Image Captioning with Scene Graphs</title><link>https://ndthp.com/projects/scene-graph-captioning/</link><pubDate>Fri, 01 May 2026 00:00:00 +0000</pubDate><guid>https://ndthp.com/projects/scene-graph-captioning/</guid><description>End-to-end image captioning pipeline conditioned on structured scene graph triples. BLEU-4 +11.4% over an image-only baseline.</description></item><item><title>QLoRA Fine-Tuning for Structured JSON Extraction</title><link>https://ndthp.com/projects/qlora-food-extraction/</link><pubDate>Fri, 01 May 2026 00:00:00 +0000</pubDate><guid>https://ndthp.com/projects/qlora-food-extraction/</guid><description>Fine-tuning a 3B instruct model with QLoRA to produce strict JSON from free-form text — on a single consumer GPU. Exact-match accuracy 0.258 → 0.624.</description></item><item><title>LLM Benchmark Evaluation: Multi-Agent Discussion Framework</title><link>https://ndthp.com/projects/llm-multiagent-benchmark/</link><pubDate>Thu, 01 May 2025 00:00:00 +0000</pubDate><guid>https://ndthp.com/projects/llm-multiagent-benchmark/</guid><description>Rigorous evaluation of a 4-agent discussion pipeline on 7,661 benchmark questions. The framework decreased accuracy on all three benchmarks — an instructive negative result.</description></item><item><title>CNN Image Super-Resolution (4× Upscaling)</title><link>https://ndthp.com/projects/cnn-image-upscaling/</link><pubDate>Sun, 01 Dec 2024 00:00:00 +0000</pubDate><guid>https://ndthp.com/projects/cnn-image-upscaling/</guid><description>Comparison of three CNN architectures for 4× single-image super-resolution on the FFHQ dataset. Evaluated with PSNR and SSIM.</description></item></channel></rss>