SS
Projects

AI Systems in Production

Each project includes measurable impact metrics — because AI engineering isn't about demos, it's about results.

75%VRAM Reduction
UnslothPEFTQLoRARAGSFT

Fine-tuning Qwen2.5-3B-Instruct

Fine-tuned Qwen2.5-3B-Instruct under tight hardware limits via 4-bit QLoRA (Quantized Low-Rank Adaptation)and paged optimizers to power a memory-efficient CRAG pipeline, optimizing structured verification logic and hallucination-free generation.

SourceLive DemoBlogResearch Paper
3.2xTraining Speedup
Sentence TransformerSparse-matrixLightGBMFastAPI

Anime-Recommender System

Built a production-grade 2-stage recommender system separating candidate retrieval and deep re-ranking. Designed using industry methodologies from YouTube’s seminal deep learning recommendation paper to optimize latency and relevance.

SourceLive DemoBlogResearch Paper