Sarfraz Javed

-Developed an agentic AI chatbot using AWS Bedrock and LangGraph's ReAct framework for autonomous customer service interactions -Built intelligent agent orchestration that dynamically selects appropriate tools (Product API, Order API, Knowledge Base) based on user intent - Designed autonomous escalation system with SNS integration that intelligently determines when to route customers to human agents - Implemented conversation persistence using DynamoDB with dynamic session management and context retention - Integrated AWS Bedrock Knowledge Base RAG integration for policy queries with intelligent document retrieval and citation

• Handled end to end data science solution delivery for one of UNICEF flagship projects, YouthHub, an app design to cater 40 million youths with opportunities related to jobs, skills, and volunteering opportunities. • Developed an advanced search and retrieval system in Python using k-NN, BERT, and ElasticSearch to recommend and filter opportunities based on user profiles. • Collaborated with cross-functional stakeholders, including product and engineering teams, to enhance features. Integrated search queries with GPT-3.5 for intent recognition, leading to a 20% increase in user engagement and a session abandonment rate reduction from 0.5 to 0.36 within four months. Also improved the F1 score from 0.52 to 0.81 (a 55% increase) based on 1,000 search/profile queries. • Conducted deep-dive NLP analysis on customer behavior, identifying key trends that improved product retention by 20%. • Worked on a fraud and compliance engine using GraphDB to detect fraudulent warranty claims, helping a client save $3 million through a rule-based detection system. • Designed a time series data quality assessment system using Change Point Detection via Binary Segmentation. Validated detected changes with ARIMA-based forecasting, ensuring accurate anomaly detection and reliable data for downstream analytics. Key Skills: Python, NLP, LLM, NoSQL, SQL, RESTful APIs, spaCy, NLTK, Time Series Analysis, Hugging Face Models, JWT, SonarQube.

• Performed Data Analysis, Insights generation and recommendations for one of India's largest FMCG company. • Conducted Topic Modelling using NMF and LDA, preparing FAQs to support the Customer Response Team. • Performed Descriptive Analysis for an airport’s duty-free customers to identify premium customers, leading to a targeted upsell/cross-sell campaign that boosted sales by 20%. • Descriptive (RFM) analysis of Online Grocery Store Data to determine most frequent bought Product by Customer, Geography etc. • Results: This helps to determine the Premium customers. A targeted campaign was launched based on insights with an objective to Upsell and cross sell. This led to an increase in sale by 20 %. • Analysed CRM data for a major airport, building a Naive Bayes Model to predict the skill set required for complaint resolution, reducing resolution time by 25% and lowering call bounce rates by 10%. • Results: The complain resolution time was reduced from 72 hours to 54 hours with reduction in call bounce rate to 15 % (previous 25%).

Developed daily and hourly stock price forecasting models of Fortune 500 companies listed on New York Stock Exchange (NYSE). Performed thorough Exploratory Data Analysis to identify Trend, Seasonality and Cycle. To get better accuracy and stable results, have compared models like Auto ARIMA ( MAPE ~70%), Holt Winter (MAPE ~50%) , Prophet ( MAPE ~50%) and Hybrid Machine Learning Model ( MAPE ~20%). Worked on Computer Vision support project based on YOLO model.

Assisted Students in Economics Commentary and Extended Essays which involve Data Exploration using Descriptive Statistics and key Economics concepts. Taught Economics, Statistics, Mathematics and Business Studies to IB and IGCSE students.

Economics and Statistics

Economics and Statistics