Usman Jadoon Logo Get in touch

/ RESUME

Usman Jadoon

AI Engineer

Atlanta, Georgia usmanzkj@gmail.com

/ PROFILE SUMMARY

AI Engineer with 3+ years of experience spanning enterprise data analytics and AI engineering, specializing in Agentic AI, Multi-Agent Systems, LLM applications, RAG, and Python development. Experienced in designing and deploying end-to-end AI platforms featuring multi-agent orchestration, dynamic cost-aware LLM routing, hybrid retrieval, Text-to-SQL, autonomous data analysis, tool and function calling, AI safety guardrails, and automated evaluation. Combines expertise in machine learning, data science, statistical analytics, and cloud technologies to engineer reliable, production-oriented AI systems that improve analytical efficiency, system performance, and data-driven decision-making.

/ EDUCATION & CREDENTIALS

M.S. in Data Science and Analytics GPA 3.96

Georgia State University, Atlanta, GA

Aug 2022 to Dec 2023

B.B.A. in Computer Information Systems GPA 3.36

Georgia State University, Atlanta, GA

Aug 2014 to Dec 2018

Technical Expertise & Core Skills

AI & LLM Systems

LLM APPLICATIONS GENERATIVE AI RAG AGENTIC AI MULTI-AGENT SYSTEMS LLM ROUTING AI ORCHESTRATION TOOL CALLING FUNCTION CALLING PROMPT ENGINEERING CONVERSATION MEMORY STRUCTURED OUTPUTS ZERO-SHOT INTENT CLASSIFICATION CRITIC FACT VALIDATION HALLUCINATION PREVENTION

Programming & Backend Engineering

PYTHON SQL FASTAPI FLASK REST APIs WEBSOCKETS ASYNC PROCESSING BACKEND SERVICES

Data Analytics & AI Data Analysis

TEXT-TO-SQL STATISTICAL ANALYSIS ANOMALY DETECTION EXPLORATORY DATA ANALYSIS SQL ANALYTICS BUSINESS INTELLIGENCE EXECUTIVE REPORTING

Machine Learning & NLP

SCIKIT-LEARN TENSORFLOW PYTORCH HUGGING FACE TRANSFORMERS NLP T5 RANDOM FOREST REGRESSION CLASSIFICATION CLUSTERING SARIMA

Data Engineering & Processing

PANDAS NUMPY PYSPARK ETL DATA PIPELINES DATA VALIDATION ALTERYX

Retrieval & Vector Search

CROSS-MODAL RAG HYBRID RETRIEVAL FAISS BM25 CHROMA VECTOR DATABASES VECTOR EMBEDDINGS SIMILARITY SEARCH RRF RERANKING MULTI-QUERY EXPANSION

AI Evaluation & Observability

AI EVALUATION MODEL EVALUATION RAGAS ROUGE LATENCY OPTIMIZATION LLM OBSERVABILITY AGENT TRACING OPENTELEMETRY BENCHMARKING

AI Security & SQL Safety

AST QUERY PARSING SQL SECURITY READ-ONLY SQL ENFORCEMENT SCHEMA-AWARE QUERYING PII REDACTION PROMPT INJECTION SANITIZATION HALLUCINATION DETECTION

Cloud & Data Platforms

AZURE ADLS GCP BIGQUERY CLOUD DEPLOYMENT POSTGRESQL SQLITE MYSQL MONGODB NEO4J

AI Frameworks & Workflow Automation

LANGCHAIN LANGGRAPH N8N WORKFLOW AUTOMATION

Development Tools & Version Control

GIT GITHUB DOCKER

LLM Providers & Local Inference

OPENAI API GOOGLE GEMINI API OLLAMA LLAMA 3.2

Data Visualization & Reporting

MATPLOTLIB PLOTLY PLOTLY.JS TABLEAU POWER BI REPORTLAB OPENPYXL

Personal Projects

AI Engineer

Feb 2024 to Present

Autonomous AI Data Analyst Platform & Multi-Agent Investigation Engine

Built a production-oriented autonomous AI data analyst platform that combines multi-agent orchestration, Text-to-SQL analytics, statistical analysis, cross-modal RAG, LLM routing, and automated executive reporting to investigate complex enterprise business questions.

  • Engineered an autonomous multi-agent analytics platform using Python, FastAPI, and LangGraph, orchestrating 8 specialized AI agents for zero-shot intent classification, Text-to-SQL generation, statistical anomaly detection, and cross-modal Vector RAG across 581,718+ (15m+ development) structured business records with sub-300ms pipeline latency.
  • Architected a fault-tolerant 4-tier LLM routing architecture spanning local Ollama (Llama 3.2), OpenAI GPT-4o-mini, Google Gemini, authenticated Hugging Face Spaces (Llama-3.2-3B), and offline rule-based fallbacks, dynamically routing analytical workloads across local, cloud, and offline inference paths.
  • Built an AST-validated Text-to-SQL pipeline with 100% read-only parse-tree enforcement, schema-aware query generation, and PostgreSQL/SQLite execution across 5 core business entities, enforcing database access guardrails for LLM-generated queries.
  • Developed a cross-modal RAG investigation engine linking quantitative SQL results with unstructured operational incident logs through hybrid vector retrieval, successfully connecting Causal Event IDs such as SC-2025-003 and PL-2025-001 with 100% Recall@5.
  • Implemented a Critic Fact Validation Agent that independently audits LLM-generated claims against raw database execution results and retrieved evidence, using bounded retry loops and evidence validation to reduce hallucinations and improve the reliability of final analytical responses.
  • Developed an automated executive deliverable compiler that transforms autonomous analytical findings into interactive Plotly.js visualizations, ReportLab PDF executive reports, and openpyxl Excel workbooks, delivering multi-format business intelligence directly through the analyst interface.
  • Established a 12-metric automated evaluation and observability framework tracking SQL accuracy, AST safety, RAG retrieval performance, hallucination detection, token usage, latency, and agent execution traces, achieving a 100% pass rate across defined target scenarios.
  • Built a full-stack ChatGPT-style analytics SPA using FastAPI, WebSockets, asynchronous streaming, persistent chat sessions, benchmark/history navigation, and interactive system architecture tracing, providing an end-to-end interface for autonomous enterprise data investigation and analysis.
PYTHON FASTAPI LANGGRAPH MULTI-AGENT SYSTEMS LLM ROUTING OLLAMA LLAMA 3.2 OPENAI API GOOGLE GEMINI API HUGGING FACE TEXT-TO-SQL AST QUERY PARSING SQL SECURITY POSTGRESQL SQLITE VECTOR RAG HYBRID SEARCH STATISTICAL ANOMALY DETECTION CRITIC FACT VALIDATION AI EVALUATION

AI Engineer

Feb 2024 to Present

Enterprise Multi-Agent RAG Platform

Built a production-oriented AI platform that combines multi-agent orchestration, enterprise knowledge retrieval, data analytics, and workflow automation to support complex business tasks.

  • Engineered and deployed a 12-layer Agentic AI, Multi-Agent RAG, and workflow automation platform using Python and FastAPI on Render, orchestrating specialized AI agents for hybrid retrieval across 15,000+ document chunks, AST-guarded Text-to-SQL analytics across 6 database schemas, structured tool/function calling, and enterprise REST API actions.
  • Improved overall RAGAS benchmark performance by 35.4% (0.5648 → 0.7650) through Multi-Query Expansion, Dense (FAISS) + Sparse (BM25) Hybrid Retrieval, and Reciprocal Rank Fusion (RRF) Reranking, driving +151.7% Context Recall and +83.3% Context Precision.
  • Engineered a cost-aware Dynamic LLM Routing and Agent Orchestration layer that analyzes task complexity and token volume to route requests across Claude, Gemini Flash/Pro, and GPT-4o, while implementing sub-millisecond (<1ms) semantic caching with FAISS vector similarity (≥0.92 cosine threshold) to reduce latency and inference costs.
  • Implemented AI safety and security guardrails including PII redaction, prompt-injection sanitization, AST-based SQL parse-tree validation enforcing read-only execution, schema allowlists, and an automated Validation Agent with claim verification and auto-retry loops to mitigate hallucinations and improve response reliability.
  • Architected end-to-end AI workflow automation and production observability incorporating AI-generated reports, ReportLab PDF compilation, structured tool/function calling, automated n8n workflows, SMTP notifications, conversation memory, state management, and OpenTelemetry distributed tracing for real-time agent execution monitoring.
PYTHON FASTAPI AGENTIC AI MULTI-AGENT RAG LLM ROUTING WORD EMBEDDING SEMANTIC CACHING RETRIEVAL-AUGMENTED GENERATION (RAG) SIMILARITY SEARCH FAISS BM25 RAGAS TEXT-TO-SQL N8N OPENTELEMETRY CONVERSATION MEMORY HYBRID RETRIEVAL VECTOR EMBEDDING FUNCTION CALLING TOOL DESIGN AI SAFETY AI EVALUATION DATA ANALYTICS RENDER

Professional Experience

Data Analyst

May 2020 to Feb 2021

Tata Consultancy Services · Microsoft Client · Redmond, WA

  • Analyzed 1M+ employee access logs stored in Azure Data Lake Storage (ADLS) using Python and SQL to identify abnormal access patterns, reducing GDPR compliance risk exposure by 20% through data-driven monitoring.
  • Developed statistical anomaly detection methods using historical baselines, threshold analysis, and behavioral comparisons to identify potential security risks and improve access auditing accuracy.
  • Performed risk-based data analysis and compliance reporting on ADLS-hosted enterprise data to support governance initiatives, anomaly investigation, and decision-making across large-scale employee access systems.
PYTHON SQL AZURE ADLS ANOMALY DETECTION GDPR

Data Analyst

Mar 2019 to Apr 2020

Tata Consultancy Services · PricewaterhouseCoopers Client · Tampa, FL

  • Analyzed enterprise data storage and usage patterns using Google Cloud BigQuery, SQL, Alteryx, and Tableau to identify inefficiencies and support data-driven resource optimization.
  • Engineered automated ETL and data validation workflows using Alteryx, SQL, and BigQuery, eliminating 40,000+ manual processing hours annually across 2,000 employees through workflow automation.
  • Developed interactive Tableau dashboards and executive reporting solutions using BigQuery data to translate complex operational data into actionable business insights.
  • Improved operational efficiency by 40%+ by identifying workflow bottlenecks, optimizing reporting processes, and supporting data strategy initiatives.
SQL BIGQUERY ALTERYX TABLEAU ETL AUTOMATION

Data Analyst Intern

Aug 2018 to Dec 2018

Fulton County Government Center · Atlanta, GA

  • Analyzed criminal justice data to uncover operational and demographic trends, improve data accessibility, and support evidence-based decision-making for government stakeholders.
  • Analyzed criminal justice datasets using R, SQL, and Tableau to identify seasonal, operational, and demographic factors influencing jail population trends.
  • Designed ETL workflows and centralized data warehouse structures, reducing data redundancy by 14% and improving accessibility of analytical datasets.
  • Performed exploratory data analysis, statistical testing, and data visualization to uncover operational inefficiencies and support evidence-based decision-making.
  • Developed interactive dashboards and analytical reports for government stakeholders, transforming raw datasets into actionable recommendations.
R SQL TABLEAU ETL

Academic Projects

E-Commerce Chatbot

AI & LLM Application Development

LLM & NLP · GPT-3.5 · LangChain · Flask · Chroma

  • Built a Retrieval-Augmented Generation (RAG) pipeline using GPT-3.5 and Chroma Vector Database to answer domain-specific e-commerce queries.
  • Embedded ~5,000 FAQ documents and evaluated performance on 200 test queries, achieving ~80% relevance accuracy.
  • Reduced response latency by ~20% through optimized vector retrieval and prompt engineering.
  • Developed and deployed a Flask-based web interface to serve model responses in real time.
GPT-3.5 RAG LANGCHAIN CHROMA NLP FLASK

Airbnb Price Prediction

Machine Learning & Predictive Analytics

Time Series & Machine Learning · SARIMA · Scikit-learn · Flask

  • Performed exploratory data analysis and feature engineering on ~10,000 Airbnb listings.
  • Built and compared SARIMA, Linear Regression, and Random Forest models to evaluate predictive performance.
  • Performed k-fold cross-validation and hyperparameter tuning using GridSearchCV to optimize model performance.
  • Random Forest achieved the best performance, capturing non-linear relationships across features with R² ~0.78.
  • Deployed the model through a Flask application, enabling automated price predictions from uploaded CSV datasets.
MACHINE LEARNING SARIMA RANDOM FOREST SCIKIT-LEARN FLASK

News Text Summarization Using T5 Transformer

Natural Language Processing & Transformers

NLP · T5 · HuggingFace Transformers · PyTorch

  • Fine-tuned a T5 Transformer model using HuggingFace Transformers and PyTorch for abstractive news summarization.
  • Preprocessed ~1,500 articles using text cleaning, normalization, and HuggingFace tokenization.
  • Evaluated model performance using ROUGE metrics, achieving ROUGE-1: 0.42, ROUGE-2: 0.28, and ROUGE-L: 0.38.
  • Optimized training through hyperparameter tuning and validation-based model selection.
NLP TRANSFORMERS T5 HUGGINGFACE PYTORCH