Portfolio · Vol. I · 2026 AI Engineering & Machine Learning

Prajwal Sathyanarayana

Building and shipping AI/ML powered products end to end —
LLM systems, agentic workflows, and models at scale.

|
View My Work ↓ Download Resume Get In Touch →
01

About Me

I build AI systems
that ship.

I'm an AI Engineer pursuing my Master's in Data Science at the University of Arizona (GPA 3.9), with 3+ years of experience building and shipping AI/ML powered products end to end — including LLM-based systems and agentic workflows.

Strong working understanding of prompt engineering, tool calling, retrieval and context construction, memory usage, and model evaluation. My work spans document intelligence pipelines, credit risk modeling, RL experiment infrastructure, and revenue forecasting automation.

Currently based in Tucson, AZ, open to full-time AI Engineer, ML Engineer, and Applied Scientist roles.

LLM Systems Agentic Workflows RAG & Retrieval Prompt Engineering AWS Model Evaluation
Contact & Details
LocationTucson, AZ
DegreeM.S. Data Science
UniversityUniversity of Arizona
Phone520-***-**** 🔒
Emailp·······@arizona.edu 🔒
GitHubPRAJWAL2108
ORCID0009-0007-0939-7736
StatusAvailable
Send a Message →
02

Experience

Jan 2026 – Present EdTech

AI Engineer Capstone

Axio Education · INFO 698

  • Engineered an async FastAPI document intelligence pipeline using PyMuPDF and pdfplumber, achieving ≥88% table extraction accuracy and <2s median processing time across 4 document modalities.
  • Integrated Gemini 1.5 Vision multimodal LLM to generate anchored margin feedback with per-region confidence scoring, automating evaluation of 10+ questions per submission at ≥87% model confidence.
  • Built a Gemini-powered document classification layer that auto-routes submissions to the correct grading pipeline, eliminating manual classification for 100% of self-contained assignment formats.
FastAPIGemini 1.5 VisionPyMuPDFLLMPython
Aug 2025 – Dec 2025 Healthcare

Data Analyst Externship

Banner Health · Eller Partnership Office, University of Arizona

  • Owned end-to-end design of AI-first workflows for complex knowledge work by building Python and SQL pipelines to extract, transform, and load encounter and procedure data, improving data accessibility by 35%.
  • Designed reusable AI capabilities and evaluation pipelines by translating messy stakeholder needs into clear specs, prompts, and expected behaviors, enabling standardized reporting structures.
  • Defined clear success metrics for AI behavior and automated reporting workflows using Python, reducing manual reconciliation effort by 40% while improving reliability of downstream surgical volume insights.
PythonSQLAI WorkflowsETLEvaluation Pipelines
Jan 2025 – Jul 2025 Research

Graduate Student Researcher

University of Arizona — ACT Lab

  • Designed and iterated multi-step agentic workflows for reinforcement learning experiments, including planning, tool invocation, memory usage, and refinement loops to support large-scale simulations.
  • Built LLM-adjacent data analysis pipelines in Python to debug, compare variants, and analyze real interaction logs from high-frequency simulations, improving experiment turnaround time by 30%.
  • Implemented model output evaluations and monitoring strategies to detect common failure modes, maintaining accuracy and reproducibility across reinforcement learning pipelines.
Reinforcement LearningAgentic WorkflowsLLMPythonDocker
Nov 2023 – Jul 2024 FinTech

Data Scientist

Cognizant · Jul 2022 – Jul 2024

  • Addressed high data processing overhead across multi-source financial datasets by architecting scalable SQL and Python pipelines, reducing cloud compute and data preparation costs by 25%.
  • Resolved costly false-positive default predictions by translating ambiguous requirements into strict engineering specs, improving detection accuracy by 18% and minimizing financial risk exposure.
  • Targeted excessive AWS resource consumption from inefficient model inference by automating pipelines via SageMaker and APIs, cutting latency by 30% and lowering cloud infrastructure spend by 20%.
AWS SageMakerCredit RiskPythonSQLAPIs
Jul 2022 – Oct 2023 Marketing

Jr. Data Scientist

Cognizant

  • Tackled high labor costs of manual revenue forecasting across business units by designing agent-like analytical workflows with tool-calling logic, reducing operational forecasting costs by 40%.
  • Mitigated expensive administrative bottlenecks in tracking system metrics by developing automated Python dashboards, improving operational efficiency by 35% and heavily reducing reporting overhead.
  • Prevented wasted operational spend on faulty A/B tests by implementing rigorous data quality checks and evaluation pipelines, stopping costly deployment of underperforming variants.
ForecastingAWS SageMakerA/B TestingPython Dashboards
03

Skills

Discipline I

AI & LLM Systems

  • Large Language Models (LLMs) & RAG
  • Vector Search & Embeddings
  • Prompt Engineering & Tool Calling
  • Agentic Workflows & Memory Management
  • Retrieval & Context Construction
  • Model Evaluation & Monitoring
Discipline II

Languages & Libraries

  • Python · SQL · C · Java · C++
  • Pandas · NumPy · Scikit-Learn
  • LightGBM · XGBoost · TensorFlow · PyTorch
  • FastAPI · REST APIs
  • Data Pipelines, ETL & A/B Testing
Discipline III

Cloud & DevOps

  • AWS SageMaker · Lambda · RDS
  • Docker & Kubernetes
  • Terraform & Infrastructure as Code
  • CI/CD & GitHub Actions
  • PostgreSQL & Version Control
Discipline IV

Product & Collaboration

  • Stakeholder Requirement Translation
  • Clear Specs & Success Metrics
  • AI-first Workflow Design
  • Cross-functional Engineering Collab
  • Fast Iteration & Edge Case Handling
04

Projects

I.

Fare Estimation for Manhattan Yellow and Green Taxis

Scalable data pipelines in PostgreSQL and Python to extract, transform, and load 3 years of NYC TLC data, enabling reliable fare prediction analytics for Yellow and Green taxis.

  • Cloud-native FastAPI deployment improving dataset accessibility
  • Automated data processing workflows for large complex datasets
PostgreSQLPythonFastAPIETL
II.

EPL Weather & Match Outcome Analysis

Interactive dashboards integrating English Premier League weather, odds, and match data for analysts and stakeholders — increasing data clarity by 40%.

  • Interactive visualizations for complex multi-source datasets
  • Automated reporting with consistent metric definitions across views
PythonDashboardsData VizAnalytics
III.

Predictive Modeling for Financial Markets

Structured data pipelines in Python to process historical market data for regression-based forecasting models with validation checks and performance monitoring.

  • High data accuracy and reproducibility across models
  • Performance metric monitoring and validation pipelines
PythonRegressionForecastingML
05

Publications

ORCID 0009-0007-0939-7736 → Google Scholar View all citations →
15Citations
/
2h-index
/
5Papers
IEEE
×6 cited

Automatic Medication Dispensing System using Machine Learning, Internet of Things and Cloud Computing

BN Hiremath, K Chavhan, NJ Johnson, P Monika, S Prajwal, TJ Reddy

2022 International Conference on Disruptive Technologies for Multi-Disciplinary Research and Applications (CENTCON) · Dec 2022

A smart medication dispensing system combining machine learning, IoT, and cloud computing to automate drug delivery, reduce dispensing errors, and improve patient adherence.

Read on IEEE Xplore → IoTCloudML
IEEE
×3 cited

Social Interaction & Service Robot for Hospital Wards

G Bharath, NJ Johnson, P Monika, S Prajwal, S Aishwarya

2022 IEEE International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE) · Apr 2022

Design and implementation of a service robot capable of social interaction in hospital environments, reducing human contact during high-risk scenarios and improving patient monitoring.

Read on IEEE Xplore → RoboticsHealthcare AI
IEEE
×2 cited

Analysis of Technologies used in Automatic Medicine Dispensing Systems

K Chavhan, NJ Johnson, P Monika, S Prajwal, TJ Reddy

2022 IEEE International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE) · Apr 2022

A comprehensive comparative analysis of technologies employed in automated medicine dispensing systems, evaluating accuracy, reliability, and scalability across approaches.

Read on IEEE Xplore → Medical SystemsAutomation
IEEE
×2 cited

Automatic Exam Answer Checker using Optical Character Recognition and Sentence Embedding

NS Yerramilli, NJ Johnson, P Monika, S Prajwal

2021 International Conference on Disruptive Technologies for Multi-Disciplinary Research and Applications (CENTCON) · Nov 2021

An automated grading system leveraging OCR to digitize handwritten exam responses and sentence embeddings to semantically compare answers against model solutions.

Read on IEEE Xplore → NLPOCRDeep Learning
IEEE
×2 cited

Navigation Systems Using A*

NS Yerramilli, NJ Johnson, OSY Reddy, S Prajwal

2021 International Conference on Recent Trends on Electronics, Information, Communication & Technology (RTEICT) · Aug 2021

Implementation and evaluation of the A* pathfinding algorithm for intelligent navigation systems, exploring heuristic optimizations and real-world applicability in dynamic environments.

Read on IEEE Xplore → AlgorithmsNavigationAI
06

Education

Aug 2024 – May 2026

Master of Science

Data Science

University of Arizona

Tucson, Arizona · GPA 3.9

In Progress
Aug 2018 – Jul 2022

Bachelor of Engineering

Computer Science

Dayananda Sagar University

Bengaluru, India · GPA 3.6

Completed