Projects — Shriansh Jain

Edge-Native LLM Inference

Bachelor's thesis researching heterogeneous compute strategies for LLM inference on NPUs using VitisAI and RyzenAI, with model quantization and local RAG system integration.

VitisAI RyzenAI NPU FastFlowLM RAG

Cloud-Native RAG System (GCP)

Scalable document chat system using Cloud Run, Cloud Build, and API Gateway. Containerized with Docker and load-balanced with NGINX.

GCP Cloud Run Docker NGINX RAG

FlounderNet - Crowd Counting

Implemented efficient architecture for crowd counting with multi-modal data fusion to enhance model representation and performance.

Computer Vision PyTorch Crowd Counting

Multiple RAG Implementations

Worked on multiple RAG strategies using langchain, and langsmith as a proof of concept.

LangChain LangSmith Vector Search RAG

HyDE RAG

Implemented Hypothetical Document Embeddings (HyDE) strategy for RAG using LanceDB for efficient vector search and LangChain for orchestration.

HyDE LanceDB Vector Search LangChain

AMCTS - Temporal Pathfinding

Implemented and deployed Adaptive Monte Carlo Tree Search algorithm for temporal path discovery in dynamic graphs.

AMCTS Pathfinding Python

Pix2Pix - Satellite to Map

Trained Conditional GAN for image-to-image translation, mapping satellite imagery to map routes with optimized generator/discriminator loss functions.

GAN Pix2Pix PyTorch Computer Vision

3D Image Encryption with GNNs

Training encryption/decryption models using PyTorch Geometric, extending the framework to work with 3D models.

GNN PyTorch Geometric Encryption

IPv4 Fragmentation Simulator

Simulates IPv4 packet fragmentation and routing using Dijkstra's algorithm. Educational tool for understanding IP networking internals.

C Networking Docker Dijkstra