• All Projects
  • AI
  • DL
  • NLP
  • UX
  • Backend / Systems
Modular RAG

Modular RAG

Flow-based RAG framework — hybrid retrieval (BM25 + FAISS, RRF fusion, MMR reranking), explicit wiring, vendor isolation, observability-first architecture.

Retrieval evaluation framework included (Hits@k, MRR, NDCG).


FastAPI Hybrid retrieval FAISS LLM
Plate2Recipe

Plate2Recipe – Food Image to Recipe Generation

Two-stage multimodal pipeline: ViT ingredient recognition → GPT-2/LSTM recipe generation.

Key finding: lower training loss (100k samples) produced worse outputs than a smaller, better-tuned run (10k samples) — quality ≠ loss.


PyTorch ViT GPT-2 LSTM
RAG v1

RAG System with Guardrails

RAG v1 — BM25 + FAISS hybrid retrieval, RRF + MMR, citation-aware answering. Led to Modular RAG architecture.


FastAPI LLM FAISS
CRM & Ticketing System

CRM & Ticketing System

FastAPI backend with Docker, RBAC, and a deployed HuggingFace Spaces demo.


FastAPI Docker RBAC Gradio UI
FoodDB — Local Nutrition Facts Database

FoodDB

Local-first nutrition facts catalogue — SQLite backend, Typer CLI, and Streamlit UI sharing one service layer. Regex parser + swappable OCR backends (Tesseract / Ollama vision).


Python SQLite Streamlit OCR
ProductLens

ProductLens

LLM-driven product comparison tool that ranks products based on user priorities.


GenAI LLM OpenAI Agents SDK Gradio UI
Professional TwinBot AI

Professional TwinBot AI

AI-powered resume-grounded assistant: tool-style prompting + retrieval patterns for accurate, personalized responses.


GenAI LLM Chatbot Gradio UI
Memory-augmented QA System

Memory-augmented QA System

Attention-based QA over structured knowledge using memory networks.


PyTorch NumPy Tokenization attention/memory networks
Notification System Redesign

Notification System Redesign

User-centered redesign of Discord's notification experience.


UX Research Prototyping Usability Testing