Currently building an AI assistants startup.● Backend: Supabase DB, Auth, Storage, Webhooks and Edge Functions. PostgreSQL. Deno. WhatsApp API.● Frontend: React.js, Next.js, Tailwind.● AI: OpenAI API, prompt engineering, function calling.

● Created an assistant that generated Solana program interfaces, based on smart contract descriptions, utilizing function calls to lint and refine the interfaces.● Implemented prompt test-driven development using promptfoo.● Engineered prompts to generate program interfaces directly from Solana Rust source code.

● Development of a ChatGPT based chatbot using Firebase, Node.js and TypeScript. Heavy use of ChatGPT function calls. Automated testing with Jest and CI/CD workflows.● Retrieval augmented generation (RAG) using Sentence Transformers for embeddings and Chroma vector database.

● NLU model development for voice bot intent classification: Extracted data from BigQuery. Prepared data using Pandas. Analyzed data for bot intent design. Assisted dataset annotation with ChatGPT. Trained and tested models using HuggingFace. Did ML tracking, conducted experiments and compared transformer-based models. Exported models to ONNX format. Utilized DVC for model versioning.● Implemented an end-to-end pipeline with DVC and Papermill for the NLU model, from dataset generation to model deployment. Established a feedback loop using Jira API, getting annotations from reported issues, checking for solved issues and updating those automatically, sending reports to Slack. This automation reduced the manual work of two–three days to zero, increasing model update frequency.● Took ownership of the NLU service project and made refactoring and improvements. Asynchronous Python code. ONNX-based model serving via a gRPC API. ChatGPT integration for specific tasks. Incorporated GitHub actions for testing and deployment. Managed configuration using Consul. Implemented A/B testing for models in production. Created a monitoring dashboard using Looker. Established log-based metrics and alerts on GCP.● Contributed to the voice bot main codebase, a multi-container application written in Go.

● Management and mentorship of the data team. I created a team of 3 junior, 1 semi-senior devs from the ground up.● Designed and leaded the development of a real-time IoT data ingest pipeline, data lake and APIs in GCP using Pub/Sub, Cloud Run, Cloud Functions, Dataflow, InfluxDB, FastAPI. This pipeline replaced a legacy process that run in a single VM. Going serverless provided redundancy and robustness to the pipeline. In addition, the improved data schema and APIs reduced some queries from 60 seconds to 2 seconds.

● Designed and developed a serverless data acquisition and model training pipeline on AWS using Step Functions, Lambda, ECS, S3, MongoDB, CDK — I did it almost entirely in TypeScript. It replaced a single process that used to run on AWS EKS and took ~10 hours to complete. Due costs involved the company updated its data/model monthly; with the new pipeline this was done almost on a daily basis, as it just needed around half an hour to finish.● Developed several data services and exposed them in a Node.js web application. One of them was a document recognizer based on AWS Textract. Another was the retrieval side of a recommendation system in Elasticsearch.

● End-to-end development of data products. Time series featurization and forecasting using several estimators such as XGBoost and FB Prophet.

● Set up a data warehouse on BigQuery, using Airflow for ETL processes.● Created models using scikit-learn and deployed them as REST APIs using Flask and Docker.