Developed an adaptive music application for board games, integrating ChatGPT (GPT‑4) with function calling to parse game narratives into discrete scenes and generate personalized soundtracks.

Engineered a robust AI‑driven search pipeline combining FAISS‑based semantic vector search with keyword filtering for highly accurate, context‑aware soundtrack recommendations.

Built an automated data ingestion process on GCP (Firestore, GCS) and deployed scalable APIs (FastAPI, Docker, Cloud Run), ensuring real‑time updates and seamless integration of AI‑powered services.

Optimized a GPT‑4‑powered chatbot for real estate to handle property recommendations, negotiations, and relocation assistance, enhancing user interactions and service quality.

Enhanced response accuracy using function calling and improved chat state management with in‑memory storage and APIs, providing responsive and context‑aware conversation flows.

Developed scalable pipelines and deployed the solution with Docker and FastAPI, integrating structured and unstructured data for personalized property recommendations and seamless operational efficiency.

Optimized in‑house chatbot responses for token efficiency and relevance, leveraging DSPy, TextGrad, and GPT‑4o to refine prompt structures.

Developed a robust system for generating synthetic question‑answer pairs and complex SQL queries (AzureSQL, PostgreSQL), enabling thorough product performance testing.

Integrated FAISS, LangChain, and GPT‑4o‑based embeddings into a unified pipeline that combines structured and unstructured data to produce multi‑layered queries and advanced QA pairs.

Developed an advanced LangGraph-based chatbot using a Retrieval-Augmented Generation (RAG) architecture to process unstructured API documentation, significantly enhancing user guidance and experience.

Created and managed a ChromaDB vector database leveraging Google and Hugging Face embedding models; tested various metadata and data-cleaning approaches to boost the chatbot's accuracy and relevance.

Structured the RAG system as a DAG with discrete nodes (retrieval, generation, verification, hallucination control), deploying on GCP Cloud Run and setting up automated triggers (GCP Cloud Functions) to ensure real-time updates for newly uploaded documents.

Developed a versatile social media bot (Twitter, Reddit) leveraging OpenAI API and Gemini Pro for organic, context‑aware interactions, including press releases and topical engagement.

Enhanced adaptability and content relevance by fetching news from 50+ sources (APIs, RSS feeds, web parsing) and summarizing data with LLMs, enabling quick and insightful updates for any given stock symbol.

Dockerized and deployed the entire service on GCP (Cloud Run, Cloud Functions, BigQuery, Cloud Scheduler) to ensure efficient, scalable, and fully automated operations.

Led end-to-end Data Science initiatives from requirements analysis to production and maintenance, fostering cross-team collaboration. Highlights include developing a custom chatbot with both Word2Vec and advanced LLM capabilities, deploying GCP model endpoints and APIs in Cloud Run, and orchestrating scheduled model retraining with Kubeflow.

Developed a robust Fraud Model utilizing advanced image/title comparison and counterfeit detection techniques (image recognition, TF-IDF, lemmatization, stemming, keyword extraction) to bolster catalog security.

Constructed and managed data pipelines for daily, weekly, and monthly feature generation in BigQuery using Python, contributed to an in-house recommender system from design to API development, and enhanced internal communication tools with an HTML/JavaScript dashboard displaying real-time order amounts via FastAPI.

Acted as an ML model developer in a post-sales automation and orchestration platform development project. Segmented customers based on Salesforce platform usage attributes.

Gathered, transformed, and summarized features to define a rule-based churn algorithm to detect possible churners among customers.

Connected to the AWS VM Instance using SSH from the local machine, set up MLFlow experiment tracking records in an AWS S3 bucket, and generated experiment track reports using Prefect.

Developed an NLP model to summarize texts using claim documents to classify customer requests and forward them to the relevant department.

Summarized effort logs of employees were collected as time series data, and then future efforts were estimated for planning future employee capacity requirements.

Constructed pipelines for gathering data from various sources such as relational databases and HTML or Excel files to generate reports; these were published via Power BI.