“Curious enough to explore. Stubborn enough to ship.”

aarya@dev:~$ whoami

Open to SDE / Backend & AI/ML Opportunities

Backend Engineer

Aarya Shirsath.

B.Tech CSE @ VIT Bhopal. I build and ship full-stack AI systems end to end and put them behind a live URL, not just a notebook.

Built

An AI pilgrimage-planning platform and a fraud model scoring 6.36M transactions, plus LLM-orchestrated assistants and retrieval-augmented pipelines.

Open source

Contributor with OSCI 2026, SWOC Season 7 and Hacktoberfest 2026.

Fundamentals

Solid DSA, databases and system design behind every build.

Groq APITemple Heritage 0.9998AUC · FraudShield AI 8Deployed products Top 10%TCS CodeVita S13

View Projects→ Resume

SCROLL

01 / About

I design, build, and ship production software — not just notebooks.

I build practical software at the intersection of AI/ML, backend engineering and full-stack development, with eleven projects, eight of them live and all with public code, not just local demos.

My work spans the full lifecycle: data modeling, API design, authentication, ML pipelines and deployment.

Recent builds

AI pilgrimage plannerGroq-driven itinerary planning with Supabase auth, visitor reviews and admin moderation.
Fraud-detection serviceScores transactions at 0.9998 ROC-AUC on 6.36M PaySim records.
Clinical decision-support platformMulti-agent triage combined with drug-interaction checks.
Healthcare appointment managerLLM pre- and post-visit summaries, email notifications and Google Calendar sync.

Open source

Contributor with OSCI 2026, SWOC Season 7 and Hacktoberfest 2026.

Fundamentals

A deliberate practice of DSA in Java, operating systems, databases and system design, reflected in a top-10% global finish at TCS CodeVita Season 13.

AI / ML LLM Applications Backend Engineering Full-Stack Cloud Deployment

GitHub↗ LinkedIn↗ Email↗

0Projects built

0Contributions this year

0Transactions modeled

0CodeVita global rank

Off-script

☕Coffee — allegedly a person, mostly just caffeine

✈️Solo travel — chasing new cities alone, one boarding pass at a time

📚Reading — collecting books faster than I finish them

🎨Painting — art as therapy, no canvas judges back

🎧Music — the soundtrack to every commit I've ever made

✍️Writing — turning bugs into blog posts nobody asked for

🧘Yoga & movement — because burnout isn't a personality trait

BaseMaharashtra, India

DegreeB.Tech, CSE '27

UniversityVIT Bhopal

FocusAI / ML · Full-Stack

Live products8 deployed, public URLs

StatusOpen to SDE / Backend & AI/ML Opportunities

04 / Stack

Tools I reach for.

Python

Java

JavaScript

TypeScript

React

Next.js

Node.js

FastAPI

PostgreSQL

Supabase

Docker

Git

TensorFlow

LangChain

GitHub

Languages

Python
Java
JavaScript
TypeScript
SQL

ML

Scikit-learn · XGBoost
TensorFlow · PyTorch
Pandas · NumPy
SHAP
Computer Vision

AI Engineering

LangChain
RAG
Multi-Agent Systems
LLM Applications
Prompt Engineering
Gemini · Groq · Anthropic

Backend

FastAPI
Node.js · Express
Flask
REST APIs
Pydantic
SQLAlchemy · Prisma

Frontend

Next.js · React · TypeScript
HTML · CSS · JavaScript
Responsive UI
Streamlit

Databases

PostgreSQL
MySQL
Supabase
SQLite
ChromaDB

Deploy / Infra

AWS
Docker
Vercel
Render
Git · GitHub

Software Engineering

OOP
DSA
API Design
Authentication · RBAC
Database Design
Testing
Monitoring

Applied in

LLM Applications→Temple Heritage · MediAgent

Multi-Agent Systems→MediAgent

AI Workflow Design→MediAgent · EduGrade

Explainable ML→FraudShield

Applied Machine Learning→FraudShield

Reinforcement Learning→Lunar Landing

05 / Activity

I build. I ship. I learn.

A public record of the work behind my projects — one commit, pull request, and shipped feature at a time.

GitHub Contribution Graph — last 12 months

Aarya Shirsath's GitHub contribution graph

View GitHub profile →

Github Stats — live

–

Total ContributionsLast 12 months

–

Current Streak

–

Longest Streak

Build Streak

18Day Streak

Active build streak across featured projects.

Achievements

OSCI 2026 Contributor

Open Source Connect India

View →

Hacktoberfest 2026

Earned IRL Sticker Pack

View →

SWOC S7 Contributor

Social Winter of Code · Season 7

03 / Experience

What I've worked on, and where.

2026

Contributor

Social Winter of Code (SWOC) Season 7

2026

Contributor

Open Source Connect India (OSCI) 2026 — Powered by NexFellow

Selected for Open Source Connect India (OSCI) 2026, collaborating with maintainers on GitHub-based projects.

SecureFlow (TypeScript): added Redis caching for AI explanations (issue #1139, PR #1160); opened PR #1204 for issue #1140, graceful handling of insufficient-VRAM failures for local models, with a typed error, a fallback model chain and a clear CLI warning instead of a stack trace.
Working inside a real review workflow: feature branches, pull requests, CI checks and maintainer feedback.

May — Jul 2026

AI/ML Intern

MPOnline Internship (Under Govt. of Madhya Pradesh)

Built and evaluated 3 TensorFlow CNN classifiers for CIFAR-10 classification, LFW face recognition, and MRI brain-tumor classification, reaching 89% accuracy. Handled preprocessing, model training, evaluation, and experimentation across multiple computer-vision datasets.

GitHub — CNN Projects → Certificate →

May — Jul 2026

SDE Intern

MPOnline Internship (Under Govt. of Madhya Pradesh)

Designed and built TaskBoard, a full-stack Flask + MySQL task management system with bcrypt-secured authentication and REST APIs (GET/POST /api/tasks) for logging, tracking, and completing employee tasks through a live dashboard.

GitHub — TaskBoard → Certificate →

Sep 2026

Job Simulation

HPE Software Engineering (via Forage)

Completed HPE's virtual software engineering simulation — writing a proposal for a RESTful web service, building a basic RESTful service, implementing data upload, and creating unit tests.

Certificate →

Sep 2026

Job Simulation

Deloitte Australia — Data Analytics (via Forage)

Completed Deloitte Australia's virtual data analytics simulation — analysing a client dataset with forensic technology, building a data dashboard in Tableau, and using Excel to classify data and draw business conclusions.

Certificate →

02 / Projects

Shipped, not just trained.

01 FEATURED PROJECT

Temple Heritage

AI-assisted pilgrimage planning for India's temples — from discovery to a shareable digital passport.

Full-stack platform for exploring Indian temples and festivals, with a Groq-powered itinerary planner, gamified pilgrimage passport, and an admin dashboard for review moderation. Every AI response is validated and normalized before it reaches a user.

Stack

Next.jsTypeScriptSupabaseGroq API

Engineering

Auth & RLSAPI RoutesAI ValidationRate Limiting

Metrics

21 temples · 14 festivals12 automated test suitesLive on custom domain

GitHub → Live Demo → Case Study →

MediAgent AI

Agentic hospital triage that scores severity and routes patients before a doctor sees them.

A 3-agent LangChain pipeline (Intake, Triage, Recommendation) followed by a deterministic safety gate that guarantees emergency cases carry an explicit warning. Scores urgency 1–10 and routes patients across 14 departments, with patient-scoped RAG chat, live OpenFDA drug-interaction checks explained by an LLM, and per-run observability in the Doctor Portal.

Stack

LangChainGroq (GPT-OSS)StreamlitSQLiteOpenFDA

Engineering

Multi-Agent PipelineSeverity ScoringDoctor PortalCase Persistence

Metrics

60-case triage eval94.4% emergency recall55 tests + CI

GitHub → Live Demo → Case Study →

FraudShield AI

Real-time fraud scoring that turns a 6.36M-transaction dataset into a single API call.

XGBoost fraud-detection model trained on 6.36M PaySim transactions with a time-aware train/test split, deployed as a rate-limited FastAPI service with Pydantic-validated schemas, optional API-key auth, and per-prediction SHAP explanations. Class imbalance is handled with imbalanced-learn, and every training run also computes a business-cost-weighted decision threshold alongside the F1-optimal one.

Stack

XGBoostPythonFastAPIPydanticRenderVercel

Engineering

Feature EngineeringClass ImbalanceTime-Aware SplitSHAP ExplanationsRate LimitingModel Versioning

Metrics

0.9998 ROC-AUC · 0.9972 PR-AUC99.49% precision / 95.93% recall6.36M PaySim transactions

GitHub → Live Demo → Case Study →

Claims Agent — Claim Extraction & Classification

Extracts every factual claim from a long answer and checks each one against a fact table with live tool calls.

Python agent that reads long answers about speaker products, pulls out each factual claim sentence, and labels it supported, contradicted or unknown by calling a brand-facts lookup tool. Claim locations are computed in Python rather than trusted from the model, and every output is strictly validated. A deterministic exact-span scorer compares a v1 baseline prompt against a v2 sentence-index prompt.

Stack

PythonGemini APIFastAPIPostgreSQLPytest

Engineering

Tool-Calling LoopStrict Output ValidationProvider AbstractionResumable Eval Runs

Metrics

≤3-turn tool loop per answerv1 vs v2 prompt comparisonDeterministic exact-span scorer

GitHub →

Ledger — Nifty Research Assistant

Turns a plain-English trading question into a structured, backtested experiment.

Research prototype that extracts a testable trading experiment from natural language, flags every assumed field before running it, and backtests it against a seeded synthetic price series. A second AI call explains the result without inventing numbers.

Stack

Next.jsTypeScriptGemini APIZodTailwind

Engineering

Structured ExtractionSchema ValidationDeterministic Backtesting

Metrics

2 narrowly-scoped AI callsZod-validated at every AI boundary

GitHub → Live Demo →

Sloth Standup — Local Git Standup Bot

Writes your daily standup from your git commits, entirely on your own machine.

Command-line tool that reads recent commits, uncommitted work and open pull requests, then has an open-weight Qwen2.5 model running inside Python write a two-line standup. No Ollama, no server, no API key, and no code leaves the laptop. Works offline after a one-time model download and can run daily through a scheduled task.

Stack

PythonQwen2.5llama-cpp-pythonGitGitHub CLI

Engineering

Local InferenceGit Log ParsingConfig-Swappable ModelScheduled Runs

Metrics

No API key, zero per-run costFully offline after one-time ~1 GB downloadMonday lookback to Friday

GitHub → DEV Post →

EduGrade-AI

Automates assignment grading with a two-layer plagiarism check built in.

Virtual-classroom platform combining Google Document AI OCR with Gemini-based grading and in-place PDF annotation. Two-layer plagiarism detection escalates to handwriting comparison, running on a normalized schema with per-row caching of AI outputs.

Stack

Document AIGeminiSQLAlchemyRender

Engineering

OCR PipelinePlagiarism DetectionRBACAI-Output Caching

Metrics

2-layer similarity checkEscalates to handwriting compare

GitHub → Live Demo →

Feedants — Competition Details Screen

A full-stack assignment where nothing is hardcoded — every state is computed live from the database.

React Native + Node.js/Express/MongoDB implementation of a competition registration flow. Spots-left, registration/submission windows, and the viewer's own participation state are all derived server-side against the current time on every request, with atomic, concurrency-safe spot booking and a server-synced countdown.

Stack

React NativeNode.jsExpressMongoDB

Engineering

Atomic Concurrency ControlDouble-Registration PreventionServer-Synced Countdown

Metrics

Atomic findOneAndUpdate spot booking20s polling reconciliation

GitHub → Live Demo →

BillGuard AI — Invoice Auditor

A vision model reads the invoice; deterministic Python rules decide what gets flagged.

Turns invoice photos and PDFs into a reviewable spreadsheet for freelancers and small businesses. An open-weight vision model (local via Ollama, or any OpenAI-compatible endpoint) extracts vendor, GSTIN, dates and totals, then Python checks flag unreadable fields, invalid GSTIN formats and subtotal + tax mismatches instead of silently guessing. Built for the Hacktoberfest Open-Source AI Challenge.

Stack

PythonFastAPIOllamaPyMuPDFPillow

Engineering

Model-Agnostic BackendRule-Based ValidationRetry & Quota Fallback

Metrics

Local-first, no per-invoice API cost30-file batches, PDF + imageCSV export + unit tests

GitHub → Live Demo → DEV Post →

Autonomous Lunar Landing — PPO

A reinforcement-learning agent judged on landings, not just reward.

PPO agent trained on Gymnasium's LunarLander-v3 with Stable-Baselines3, using a custom callback that tracks safe landings, crashes, and timeouts separately rather than relying on mean reward alone. Includes 50k-step checkpointing and a deterministic evaluation harness.

Stack

Stable-Baselines3GymnasiumPyTorchTensorBoard

Engineering

Custom CallbacksCheckpointingEvaluation Harness

Metrics

Outcome breakdown: safe / crash / timeout50k-step checkpoints

GitHub →

Healthcare Appointment & Follow-up Manager

One platform, three portals — patient, doctor, and admin, kept in sync by an LLM.

Full-stack clinic platform where an LLM drafts urgency-flagged pre-visit summaries and patient-friendly post-visit notes. Automated email confirmations, Google Calendar sync via OAuth 2.0, and medication reminders round out the patient side.

Stack

Node.jsPostgreSQLPrismaGemini

Engineering

OAuth 2.0Email AutomationScheduled Reminders

Metrics

LLM summary on both sides of every visitAutomated Calendar sync

GitHub → Live Demo →

06 / Education

Grounded in fundamentals.

2023 — 2027

B.Tech, Computer Science & Engineering

VIT Bhopal University

Coursework: Data Structures & Algorithms · Operating Systems · DBMS · OOP · System Design · Machine Learning · Advanced Java.

12th Standard

Higher Secondary (12th)

Jain International School

CBSE · Science stream.

10th Standard

Secondary (10th)

Tender Care Home

CBSE.

07 / Certifications

Credentials worth listing.

TCS CodeVita Season 13

Tata Consultancy Services

Top 10% globally · 90th percentile · rank 15,033

View Certificate →

Certified Software Engineer

HackerRank

View Certificate →

AWS Cloud Training

Intellipaat

EC2 · S3 · IAM · VPC

View Certificate →

ServiceNow Virtual Internship

ServiceNow University × AICTE × SmartBridge

Flow Designer · Agentic AI · ATF

View Certificate →

Software Engineering Job Simulation

Hewlett Packard Enterprise (HPE) · via Forage

Wrote a RESTful web service proposal, built the service, implemented data upload, and added unit tests · REST APIs · Testing

View Certificate →

Data Analytics Job Simulation

Deloitte Australia · via Forage

Analysed a client dataset with forensic technology, built a Tableau dashboard, and used Excel to classify data and draw business conclusions · Tableau · Excel

View Certificate →

Bits and Bytes of Computer Networking

Coursera

TCP/IP · DNS · HTTP

View Certificate →

Marketing Analytics

NPTEL

95% · Elite

View Certificate →

Programming in Java

Vityarthi

View Certificate →