SP
Open to AI/ML Opportunities

Hi, I'm Silu Poudel

I'm an |

AI Engineer specializing in LLM systems, RAG pipelines, and Agentic AI. Building production-ready AI solutions with LangChain, FastAPI, and vector databases. Based in Nepal 🇳🇵

View My Projects Get In Touch
Scroll Down
Get To Know Me

About Me

AI Engineer — LLM Systems, RAG, and Agentic AI

I'm an AI Engineer from Nepal specializing in LLM systems, RAG pipelines, and Agentic AI.

I've built production-ready AI backends at Grinda AI, where I developed a hybrid OCR pipeline for Korean financial documents, and completed the AI Fellowship at Fuse Machine working on an NLP-based mental health chatbot.

I enjoy turning complex AI research into real-world systems that actually work.

What I Know

My Skills

Machine Learning & AI

LLM Systems RAG Pipelines Agentic AI ReAct Pattern Prompt Engineering Fine-tuning (Qwen) Model Evaluation

LLM & RAG

LangChain LangChain LCEL Ollama OpenAI API Qdrant ChromaDB FAISS HuggingFace

OCR & Computer Vision

PaddleOCR Qwen-VL GLM-4V OpenCV CNN CER/WER/TEDS Metrics

Natural Language Processing

Text Preprocessing Sentiment Analysis Emotion Detection Crisis Keyword Detection Transformers

Programming & Backend

Python FastAPI REST API Development Redis SQLAlchemy Async Processing Docker

Data & Databases

Vector Databases (Qdrant, FAISS, ChromaDB) PostgreSQL SQLite Pandas NumPy

Frontend (Working Knowledge)

React JavaScript HTML / CSS
What I've Built

My Projects

FastAPI Document RAG System

Production-ready RAG backend that ingests documents, creates vector embeddings with OpenAI, and serves answers via REST API. Uses Qdrant as vector store, Redis for caching, and PyMuPDF + LlamaParse for parsing.

Highlights

  • End-to-end ingest → embed → retrieve → respond pipeline
  • Redis caching cuts repeated query latency
FastAPI Qdrant OpenAI Redis LangChain
View Code

Mental Health Chatbot

AI chatbot for emotional support using fine-tuned Qwen models for emotion and intent classification. Uses FAISS for semantic memory, SQLite for session persistence, and real-time crisis keyword detection for escalation.

Highlights

  • Fine-tuned Qwen 0.6B embeddings + Qwen3 1.7B generation
  • Crisis detection layer with keyword-based escalation logic
Qwen FAISS SQLite Python NLP
View Code

Face Emotion Recognition System

Real-time facial emotion detection trained on FER2013 48x48 grayscale images. Built a CNN from scratch, reduced overfitting via augmentation and regularization, and integrated live webcam inference with avatar mirroring.

Highlights

  • Custom CNN trained on 35,000+ labeled images
  • Live prediction with OpenCV webcam
CNN TensorFlow OpenCV FER2013
Private Repo

Hybrid OCR Pipeline — Korean Financial Docs

Built a hybrid OCR pipeline at Grinda AI combining PaddleOCR, Qwen-VL, and GLM-4V for processing Korean financial documents. Generated synthetic training samples and benchmarked against GPT-4o and Upstage using CER, WER, and TEDS metrics.

Highlights

  • Hybrid pipeline combining classical OCR with vision-language models
  • Benchmarked against GPT-4o and Upstage on CER/WER/TEDS
PaddleOCR Qwen-VL GLM-4V FastAPI Python
Private (Company)
My Journey

Experience

AI Engineer

Grinda AI · Nov 2025 – March 2026
  • Built a hybrid OCR pipeline for Korean financial documents combining PaddleOCR, Qwen-VL, and GLM-4V.
  • Generated synthetic training samples and benchmarked against GPT-4o and Upstage using CER, WER, and TEDS metrics.
  • Developed production AI backends with FastAPI, integrating Qdrant vector store and Redis caching for low-latency retrieval.
  • Owned end-to-end pipelines from document parsing (PyMuPDF, LlamaParse) to embedding, retrieval, and response.
FastAPI PaddleOCR Qwen-VL GLM-4V Qdrant Redis

AI Fellowship Intern

Fuse Machine
  • Worked on NLP-based mental health chatbot system
  • Collaborated in AI project development
  • Applied ML model evaluation techniques
  • Built end-to-end ML pipelines from data to deployment
NLP Machine Learning Python Scikit-learn
Let's Talk

Get In Touch

Let's work together!

Interested in AI/ML collaboration, research opportunities, or building intelligent systems? Feel free to reach out. I'm always open to discussing new projects and ideas.

Location

Nepal

Website

silu.com.np