PROJECTS
Conversational AI
We developed a LLM-based conversational AI model to offer 24/7 support to our client’s customers. The pipeline we developed covers the full data pipeline: from handling client-specific data ingestion requests, customizing the model to their own data and style, to building an interactive dashboard to keep track of the user satisfaction and FAQs. Using our product, the client was able to improve the quality and speed of the response delivery, as well as cut down their cost.
- Large Language Models
Knowledge Assistant
For a California-based startup, we developed the backbone of a knowledge assistant, which offers users a wide rage of analytics over their documents. Using state-of-the-art technologies, our team delivered different features: from text and entity classification, to text summarization using Large Language Models (LLMs). It is the next step in the evolution of operating systems as you know them.
- Agglomerative Clustering
- Retrieval Augmented Generation (RAG)
- Named Entity Recognition using UniNER
- Text summarization using LLMs
- Entity Matching
- Graph Analytics
- Random Forests
- Text Classification Models
Influence Maximization on Large Social Graphs
- Graph Algorithms – Diffusion and Influence Maximization heuristics
- Image Recognition
- NLP
- Social networks analysis
Automated AI-based Stock Trading
We developed a fully automated AI-based trading platform for a NY-based hedge fund with M$ in AUM. Out team curated all aspects of the project: data ingestion – batch and real time – using multiple heterogenous sources, development of alpha generating algorithms using advanced AI and Machine Learning, and finally trading execution in real time using Bloomberg ESMX through a primary and secondary broker. Collaborated with the quant team in NYC and data science team in California.
- Python
- MongoDB
- Real-time Ingestion
- AWS
- EMSX
- B-PIPE
- NLP
- Technical Analysis
- Dash
- Pytorch
- FB Prophet
- Graph Analytics
- Sentiment Analysis
- Technical Analysis
- Deep Neural Networks
- Random Forests
- Anomaly detection
Knowledge Extraction, Representation and Exploitation (from Unstructured Data)
Our Client, an innovative AI-powered Silicon Valley startup, partnered with us for the development of their Machine Learning platform for knowledge extraction, representation and exploitation. Our team worked on problems, including but not limited to:
- Entity extraction and disambiguation from text
- Information retrieval
- RecSys
- Text pre-processing
- Topic analysis
- Exposure analysis using graph analytics
- Sentiment analysis
- Pipeline creation using AWS (EC2, EMR, Lambda, SQS, Athena, etc)
Our algorithms have been integrated into dfferent verticals served by the Client and ultimately adopted by Stanford University, Moody’s, CITI, etc.
- Python
- Apache Spark / EMR
- NoSQL DB
- AWS
- ElasticSearch
- Word Sense
- Disambiguation and Concept Extraction
- (Hierarchical) LDA
- Word(Doc)2Vec, (s)BERT, Text Embeddings
- Node2Vec and Graph embeddings
- Time-series Forecasting (Statistical, Neural-Network based)
- Genetic Algorithms for Hyperparameter Optimization
Human Posture Extraction and Analysis
Contacted by an LA-based VC, we developed a computer vision and AI driven platform for human posture extraction in real-time enabling powerful applications across the Social, Health and Fitness, and Sports verticals. The solution we developed extracts relevant keypoints (position and orientation of joints) for multiple humans being tracked, and compare them to a database of known motor acts assigning a score to each comparison – all this in real-time. Think of our approach as a Shazam for human activity recognition!
- Python
- NodeJS
- MongoDB
- GraphQL
- AWS
- Circle CI
- Computer Vision for Human Pose Extraction
- Dynamic Time Warping
- Approximate kNN
- Kalman Filtering
Robo-Trading using Technical Analysis
One of the leading European financial education platforms partnered with us to develop tools for suggesting their clients potential trading strategies using solely technical analysis. We also deployed an ad-hoc scripting language for more advanced clients to develop their own strategies based on signals from more than 30 technical indicators. The platform was deployed on AWS and is running daily on a selection of liquid stocks.
- Python
- Technical Analysis
- AWS
- Lex/Yacc
- Technical Analysis Algorithms
