This is a list of arbitrary projects and research papers I managed, supervised. All of them has been in our teams in Miras and Eveince. The NFT project was a solo project.
-
RADLLM
A platform architecture for rapid LLM application development. I designed and developed:
- Platform architecture to build ETL pipelines and embedding DB (plus mechanisms for costs savings)
- Serverless functions for LLM interactions and embedding APIs
- Backend services for prompt handling
-
Outperforming Oracle Finance for credit scoring
In this work, we designed an enterprise credit scoring product benchmarked against Oracle Finance solution that used regression models for credit prediction. After our initial EDA, we designed a new solution based on LeggoApp that exploits state machines for different customer segments and, for each customer segment, used different models comprising of SVM and Decision Trees for scoring tasks.
Credit risk rating using state machines and machine learning
- MirasText: A language model as a NLP toolkit
We crawled, labeled, and modeled text data from Twitter, Telegram, News websites, etc, to build a large model that enabled us to perform natural language processing tasks like sentiment analysis, name-entity recognition, and topic modeling at scale.
-
Other projects
Using active learning to optimize sentiment analysis
Graph-based approach for Twitter trend analysis
- NFT project (Embeddings of light): With the rise of generative models, I started some experiments on digital art, used them in Eveince branding, and published them as an NFT collection.