Building ML infrastructure, recommendation engines, and LLM agents

As a Data and AI consultant, I work with companies in leveraging their unused data effectively and applying novel technologies (AI/ML) to solve problems or enhance existing features. I typically work with companies that are just starting to explore AI solutions and are interested in exploring the possible potential before diving in too deep.We typically work on ,1. Understanding legacy data and transforming where necessary to make full use of it.2. Finding potential gaps in data and filling them.3. Applying AI/ML approaches to automate and speed up slow processes.4. Identifying new avenues using the latest technologies to drive more revenue.A few examples include,1. Chatbot extension that lets users chat with their own data (while maintaining full privacy)2. AI agent based research assistants that conduct and summarize research at scale saving 100s of hours.3. Data pipeline orchestration streamlining the data flow across multiple platforms.4. Automatic labelling of data with little human intervention.5. Recommendation systems to find similar products

Lead the Data Team at Noibu.I was involved in,- Taking a customer problem and working backwards to design a data/engineering solution.- Setting the long-term and short-term data initiatives for the company.- Building complex data pipelines to collect, store, analyze, and understand structured and unstructured data to help our business make decisions and serve our customers. - Deploying algorithms and ML models on the cloud.- Built end-end ML pipelines in Rust using ONNX and custom frameworksTools:- PostgreSQL, ElasticSearch, BigQuery, cockroachDB.- Python, Rust, PyTorch, scikit-learn, spaCy, Huggingface to implement ML/NLP approaches.- My work was in the areas of NLP, LLMs, semi-supervised learning, classical ML(Classification, Regression, Clustering, Ensemble methods etc.), distributed data processing.

Data Scientist: - I worked with cross-disciplinary teams in understanding and analyzing data.- I built Deep Neural Network models (auto-encoders) to automate processes to make faster decisions.- My work involved heavy usage of computational statistics, programming, data analysis and involved interactions with stakeholders from different disciplines (Doctors/Engineers) - My main area of focus was unsupervised learning, Deep ANNs, and NLP using Python, spaCy, fastai,PyTorch.Project Manager, Data Science:- I founded the Fri-ML organization to work collaboratively with ML experts in the Fribourg-Bern region and to teach ML concepts and the impact they have on our daily lives to the general public.- We worked collaboratively on a range of business and fun problems in computer vision, NLP and structured data using PyTorch and fastai.- As the project manager, I organized the workflow, assigned jobs, attracted/invited new participants, organized meetings and delivered lectures.

- I lead an international collaboration among scientists from the US, UK, Belgium, and Switzerland to develop a numerical model for Hysteresis in porous building materials.- I made use of probability theory, statistics and calculus to build the model.- The work was very successful and was presented at two major international conferences (Boston and Frejus). The one I presented at ICNEM 2014 in Frejus, France won a prize.