OPEN TO SOFTWARE ENGINEERING ROLES

Intelligent systems.
Built to deliver.

I’m Pratik. I build and deploy cloud applications, LLM-powered products, and agents — from the interface to the infrastructure behind them.

Explore my work
GitHub LinkedIn Résumé
FROM BUILDING TO OPERATING Cloud developmentAWS deployment · Application & data LLM engineeringHybrid retrieval · Grounded answers Agentic systemsOrchestration · Shared state · Recovery
01 / SELECTED ENGINEERING WORK

The build. The thinking.
The working system.

Cloud delivery, LLMs, and agents.
Explore the decisions behind each project.

FROM BOOKING FLOW TO CLOUD02 / AWS DEPLOYMENT
A ticket.
A complete journey.
Booking interface · Application · AWS

Train Ticketing on AWS

CLOUD & APPLICATION DEVELOPMENT

A train reservation project deployed on AWS, bringing together booking screens, checkout, and a related PHP/MySQL application.

AWSHTML / CSSPHPMySQL

ENGINEERING FOCUSCloud delivery across the interface, application, and data layers.

Build & deployment notes
The application
Route selection, fare calculation, checkout screens, and booking history form the passenger journey.
The system boundaries
The hosted-project repository contains the HTML/CSS presentation. The companion application contains PHP session handling, booking logic, and the MySQL schema.
Cloud delivery
I deployed the project on AWS. Both repositories are linked below so the interface and application implementation can be explored separately.
Hosted-project source ↗PHP / MySQL source ↗
AN ANSWER YOU CAN TRACE03 / LLM + RETRIEVAL
Better answers.
With receipts.
Reviews → Hybrid retrieval → Cited answers

GMU Housing, Grounded

RETRIEVAL-AUGMENTED GENERATION

A housing assistant that searches tenant reviews and listings, then answers questions with references to its source material.

PythonChromaDBBM25Groq

ENGINEERING FOCUSHybrid retrieval, source attribution, and explicit evaluation.

Retrieval & evaluation notes
The design decision
Combine semantic retrieval with BM25 keyword matching using reciprocal rank fusion. Embeddings capture meaning; keyword search helps retain exact apartment names.
Grounding the model
Title-prefixed chunks retain property context. Retrieved passages are numbered, and the generation prompt asks the model to cite them and decline questions the sources cannot answer.
What I evaluated
The repository includes a five-question evaluation, a hybrid-versus-semantic comparison, and analysis of a failure case where an overall housing judgment lacked balanced evidence.
Open Streamlit demo ↗Read the evaluation ↗
A PLAN THAT RESPONDS TO RESULTS04 / MULTI-TOOL AGENT
Find. Compare.
Make it yours.
Search → Compare → Style · Retry when needed

FitFindr

PLANNING, TOOLS & STATE

A fashion assistant that searches a secondhand catalog, compares prices, and uses an LLM to suggest outfits and write a fit card.

PythonGroqGradiopytest

ENGINEERING FOCUSConditional planning, shared session state, and graceful fallback.

Planning & recovery notes
The orchestration
A rule-based planning loop coordinates five tools. Search and price comparison use deterministic Python; outfit suggestions and fit-card writing use the LLM.
The failure path
When search returns nothing, the agent tries relaxed filters and records the changes. If no item is found, it offers catalog trends and stops before calling the generation tools.
The state model
One session object carries the query, selected item, price assessment, and generated output between steps. Tests and failure drills document the expected behavior.
Watch the walkthrough ↗Read the planning loop ↗
More interfaces, research, and experiments.Explore my GitHub
More from the workshop 7 PROJECTS
AI Agentic BrowserReact · Electron · GPT / Claude / Gemini Fatigue PredictionSwift · HealthKit · Machine learning SmartSplitReact · TypeScript · Firebase · OCR Adaptive ML VisualizationVTK.js · WebXR · TensorFlow.js Edulearn BoltTypeScript · Supabase · AI chat Smart Clipboard ManagerJavaScript · Manifest V3 · Local AI Speech Analytics DashboardReact · Web Audio API · Speech metrics
02 / A CLOSER LOOK AT THE ENGINEERING

A model is one part.
I build the system around it.

Inside IndustrialBriefs:
agents, data, delivery, and operations.

INDUSTRIALBRIEFS / PUBLISHING ARCHITECTUREExplore the documented system ↗
  1. 01 / DISCOVERScout

    Collect, score, and deduplicate incoming stories.

  2. 02 / GENERATEWriter

    Draft articles with GPT-4o and structured output.

  3. 03 / ENRICHSEO & context

    Add metadata, fact sheets, and related coverage.

  4. 04 / DELIVERPublisher

    Publish to Ghost and distribute through reader channels.

SHARED STATESupabase

Feeds · Queue · Clusters · Fact sheets · Published records

OPERATIONSMaintainer + Error Handler

Workflow health · Recovery · Failure classification · Alerts

01 / CLEAR RESPONSIBILITIES

Give every agent a job.

Discovery, writing, enrichment, and publishing have distinct responsibilities, connected through shared state.

02 / INSPECTABLE OUTPUTS

Keep the evidence visible.

Structured article data in the newsroom; source-linked answers and retrieval comparisons in the housing project.

03 / EXPLICIT FAILURE PATHS

Design what happens next.

Operational recovery in IndustrialBriefs, and a tested early exit when FitFindr cannot find a usable result.

CLOUD & OPERATIONSDeployment details matter, too.

Alongside the AWS ticketing project, my Oracle Cloud automation deployment package covers the operational side of running a service.

Network boundary
Caddy handles HTTPS at the edge. n8n and PostgreSQL communicate on Docker networks; the database network is internal.
Service lifecycle
Compose health checks gate service startup, persistent volumes keep application state, and configuration is supplied through environment variables.
Recovery
A database export script keeps a rolling 14-day local history. The runbook covers off-server copies, health verification, and checks before an upgrade.
Read the deployment notes
03 / THE PERSON BEHIND THE CODE

Curious by nature.
Engineer by craft.

Pratik PatilDEVELOPER · BUILDER · CONSTANT LEARNER

I like understanding how the whole thing works — and making it work better for the person on the other side of the screen.

I’m pursuing an M.S. in Computer Science at George Mason University. My work connects application development, cloud delivery, and applied AI: an AWS-hosted booking project, an agentic newsroom, and retrieval systems that make their sources visible.

I care about the decisions between the tools: when retrieval needs exact keyword matches, when an agent should stop, and how a service recovers when something fails. I use AI in my development process, and take responsibility for reviewing, testing, and understanding what ships.

M.S. Computer ScienceGeorge Mason University · 2024–2026 · GPA 3.78/4.0
Education, research & credentials

B.Tech in Computer Engineering
University of Pune · CGPA 8.51/10.0

Published research · Springer ICSCS 2023
Fruits Image Analysis using Computer Vision & Deep Learning

  • AWS Certified Cloud Practitioner
  • CodePath Intermediate Web Development · Honors
  • Oracle OCI Developer Professional
  • Oracle OCI Generative AI Professional
  • Oracle OCI Data Science Professional
  • Oracle AI Autonomous Database Professional
MY TOOLKIT

Skills connected
to working systems.

Each layer has a job.
I build the connections between them.

Interfaces & applications

Booking flows, desktop interfaces, API integrations, and the data models behind them.

React · Next.js · TypeScript
Node.js · FastAPI · Spring Boot

AI & machine learning

Hybrid search, source-grounded generation, tool orchestration, shared state, and failure analysis.

Python · Groq · ChromaDB · BM25
LLMs · RAG · Agent workflows

Cloud & data

Cloud deployment, containerized services, HTTPS, internal database networks, and recovery scripts.

AWS · Oracle Cloud · Docker Compose
PostgreSQL · MySQL · Caddy · n8n
04 / EXPERIENCE

Building systems.
Owning the details.

Building alongside teams.
Taking ownership of the details.

View full résumé
2026 — PRESENTCURRENT

AI Solutions Engineer

VisioneerIT

Building production AI and data systems with Python, PostgreSQL, containerized services, and AWS infrastructure.

2026 — PRESENT

Software Engineer Intern

GUTTS

Turning spreadsheet-heavy operations into CRM applications, Python automation, and connected workflows.

2026 — PRESENT · VOLUNTEER

Forward Deployed Engineer

Blue Horse Digital

Translating business needs into voice-AI intake flows, integrations, and clear paths for human support.

JUN — AUG 2023

Software Engineering Intern

Persistent Systems

Developed Java modules, improved SQL performance, and contributed to debugging and code reviews in Agile sprints.

APR — JUN 2023

Web Developer Intern

PHN Technology

Built Node.js and MongoDB backend features, integrated REST APIs, and improved client-side experiences.

OPEN TO OPPORTUNITIES & GOOD CONVERSATIONS

An idea. An agent.
A system. Let’s build it.

LinkedIn ↗GitHub ↗