01 / Introduction

HARSH GAVAND / FULL-STACK & AI ENGINEER

I build the interface.
And the system
behind it.

Full-stack applications, backend APIs, and applied AI workflows — from the data model and provider integration to the interface people use.

Explore selected work Get in touch

Mumbai, India · Open to Full-Stack, Backend & AI Engineering Roles (Remote / Relocation)

engineering / field notes

Interfaces → APIs → workflows → data

Typed interfaces, request state, and accessible interactions.

React · TypeScriptSee the implementation

server/quota-lifecycle.mjs

1const estimate = reservationFor(request);2const reservation = await quota.reserve(estimate.amount);3let providerDispatched = false;4let actual;5let result;6try {7  signal.throwIfAborted();8  providerDispatched = true;9  result = await generate(request, { signal });10  actual = usageCost(result.usage, estimate);11  await quota.settle(reservation, actual);12} catch (error) {13  await quota.releaseOrEnqueueRetry(reservation, {14    providerDispatched,15    actual,16  });17  if (actual === undefined) throw error;18}19return result;

Atomic quota reservation, pre-dispatch abort rollback, and durable settlement retries with dead-letter handling.

Read the quota lifecycle

4End-to-end systems 2+ yrsFull-stack & backend engineering

StackTypeScript · Node.js · Python · React · AI

02 / Selected Work

Four systems. The engineering behind them.

AI media generation, lending workflows, repository discovery, and automated news delivery.

01Applied AI & Media Pipeline

UGC Content Studio — AI Media Generation Pipeline

Live demoClerk Auth · Live AppGitHub source

The problem & product

Marketing teams need rapid product and model variations without repeated manual shoots. Engineered an end-to-end media workflow coordinating Clerk-authenticated uploads, Google AI image/video synthesis, Cloudinary CDN delivery, and PostgreSQL project and credit records.

Architecture & engineering decisions

Bound project queries to Clerk user identity through Prisma ownership predicates, keeping account and generation records related in PostgreSQL.
Coordinated Google AI generation, provider-operation polling, and Cloudinary uploads with persisted isGenerating, output URL, and error fields.
Used Prisma credit increment/decrement operations and failure-path restoration for caught generation errors, preserving project error state for the interface.

React
TypeScript
Node.js
Express
Prisma
PostgreSQL
Clerk
Cloudinary
Google AI

Engineering outcome

Persisted generated image/video URLs and project error state in PostgreSQL, with Prisma credit updates and restoration paths for caught generation failures.

Studio workspace / interface study

Workspace: Brand Shoot #14

Status: CompletedCredits: 5 / image · 10 / video

UGC project gallery image of a model with a suitcase

Generation pipeline

Input 01 / Product reference: Suitcase asset + product description
Input 02 / Model conditioning: Model reference image + generation prompt
Output / Cloudinary CDN: Generated image + short video asset

Clerk identity → Google AI → persisted media URLs

Interface reconstruction using the project’s gallery asset. Credit values show configured generation costs.

Open the product

02FinTech & Lending Workflows

LendFlow — Multi-Role Lending & Underwriting Platform

Live demoLive Multi-Role AppGitHub source

The problem & product

Lending workflows span borrower origination, lender review, and administrative oversight. Engineered a role-oriented platform with relational borrower/loan records, decimal principal and interest fields, explicit review states, and React dashboards.

Architecture & engineering decisions

Modeled borrower foreign keys, DECIMAL(10,2) principal, DECIMAL(5,2) interest, and pending / approved / rejected loan status values in Sequelize.
Implemented JWT account authentication and Express application, lookup, approval, and rejection handlers with missing-record responses.
Structured borrower, lender, and admin interfaces around application, repayment, and review data, with Chart.js visualizations for financial context.

React
Node.js
Express
Sequelize
SQL
JWT
Chart.js

Engineering outcome

Represented loan principal and interest with DECIMAL fields, linked applications to borrower records, and persisted explicit approval/rejection outcomes for lending and repayment interfaces.

Banking application landing page introducing lending and investment workflows

Open the landing page; account access is needed for borrower, lender, and admin views.Render-hosted deployment; the first load can take longer while the service wakes.

Open the product

03Developer Tools & Document Relations

DevGraph — Repository Discovery & Issue Indexing Platform

Live demoLive Explorer AppGitHub source

The problem & product

Developer collaboration depends on traversable relationships between accounts, repositories, and issues. Engineered a repository explorer backed by Mongoose document references, Express mutation handlers, and independent personal/discovery feed state.

Architecture & engineering decisions

Modeled User, Repository, and Issue references in Mongoose, using owner and issue population in repository queries to resolve related documents.
Validated required repository names and owner ObjectIds during creation; separated lookup, content updates, visibility toggles, and deletion into Express controllers.
Separated personal and suggested repository feeds from the local search query so filtering does not trigger a network round-trip per keystroke.

React
Node.js
Express
MongoDB
REST API

Engineering outcome

Resolved owner and issue relationships through Mongoose population and filtered loaded repository names locally without a network request per keystroke.

GitHub Replica account sign-in interface

Create an account or sign in to explore personal repository views.Render-hosted deployment; the first load can take longer while the service wakes.

Open the product

04AI Automation & Delivery Pipelines

AI News Automation — Ingestion-to-Email Pipeline

Local Pipeline AppGitHub source

The problem & product

Repeated article and transcript collection needs stable source identity and inspectable delivery stages. Engineered a Python ingestion-to-digest pipeline with GUID/video-ID checks, local Ollama inference, SQLAlchemy persistence, and multipart SMTP output.

Architecture & engineering decisions

Normalized articles and videos into SQLAlchemy records, checking GUIDs and video IDs before insertion to skip sources already stored.
Separated collection, local Ollama summarization, digest persistence, and email formatting; model failures return an explicit missing result instead of fabricated content.
Validated recipient and SMTP configuration before delivery and generated multipart text/plain and text/html digests with original-source links.

Python
SQLAlchemy
Docker
Ollama
Pydantic
SMTP

Engineering outcome

Skipped previously stored source identities, persisted source-linked digests, and validated recipient and SMTP configuration before preparing multipart plaintext/HTML delivery.

Email delivery format

One digest. Two formats.

Plain text

A readable summary with links back to the collected articles and transcripts.

HTML email

A greeting, introduction, article headings, formatted summaries, and source links.

text/plain + text/html · SMTP delivery

The product output is a multipart email digest. Inspect the formatter and run instructions in the repository.

Inspect the email formatter

Curated archiveEight more engineering problems.Browse repositories

Selected AI, API, retrieval, and workflow repositories
Year	Project	Built with	Key engineering feature	Links
2026	Banking Agent Harness	Python · Google ADK · LiteLLM	Deterministic transfer policy and later-turn confirmation	GitHub
2026	Hybrid RAG Documentation Assistant	Python · LangGraph · ChromaDB	BM25 + vector retrieval with Reciprocal Rank Fusion	GitHub
2026	AI Support Ticket Classifier	Python · FastAPI · LangGraph	Validated classification with redaction and review fallback	GitHub
2026	Symposium RAG · Semantic Cache	Python · FastAPI · Streamlit	Semantic answer reuse with source-page retention	GitHub
2026	Healthcare MCP & OpenAI	Python · MCP · FastMCP	MCP discovery and bounded read-only SQL tools	GitHub
2026	Multi-Agent Customer Support	Python · Streamlit · LangGraph	Separate conversational intake and approval graphs	GitHub
2026	AI Voice Adventure Assistant	Python · FastAPI · JavaScript	Async transcription → generation → speech playback	GitHub
2026	ADK Shopping Assistant	Python · Google ADK · LiteLLM	Validated cart state across four ADK agents	GitHub

Browse the complete project catalog

03 / Experience & Impact

Experience that connects the layers.

From responsive interfaces to APIs, relational data, and application delivery.

Globetrek Engineering Corporation

Nov 2025 — Sep 2026

AI-Native Full-Stack Developer

Engineered Django and React business workflows across relational models, REST contracts, interface state, and deployment, delivering features across the complete application lifecycle.
Developed Python and JavaScript APIs, asynchronous tasks, and third-party integrations to move long-running service work into explicit backend workflows.
Designed relational schemas and reusable SQL reporting queries to keep operational records consistent across application screens and reports.
Reviewed shared changes against coding standards and reusable implementation patterns, keeping collaborative delivery maintainable across frontend and backend boundaries.

Django
React
Python
JavaScript
SQL
REST APIs
AI / ML
LLMs
RAG
Docker
Grafana
k6
CI/CD

HRP Enterprises

Jan 2024 — Oct 2025

Full-Stack Developer

Engineered responsive interfaces across browser and screen-size variations, supporting 1,000+ concurrent users with no reported compatibility issues.
Integrated REST APIs across 10+ pages and optimized page delivery, reducing average load time by 30%.
Resolved 20+ interface defects through device and screen-size fixes, improving consistency across responsive layouts.

HTML
CSS
JavaScript
TypeScript
React
Node.js
Express
REST APIs
SQL
Docker
Kubernetes
AWS
GCP
CI/CD
Git

About me

Curious about the whole system.

I build full-stack applications, backend APIs, and applied AI workflows. My work connects React and TypeScript interfaces to Node.js and Python services, relational and document databases, and external model and media providers.

Education

B.Tech · Information Technology

Pillai University, Mumbai · July 2025 · CGPA 8.6

Based in

Mumbai, India · Open to relocation / remote

Hackathons & recognition

Top 5

Supervity AI Hackathon, 500+ participants, built in 24 hours

View project ↗

Top 10

OpenAI Builders Hackathon, Memory Graph AI, 500+ builders

View project ↗

Top 100

Smart India Hackathon 2025, national level, 500+ teams

04 / Architecture & Skills

Architecture before abstraction.

The tools I use across interfaces, services, AI workflows, and data — and the principles that connect them.

01 / Backend & APIs

Contracts, identity, and request admission.

Node.js
Express
Python
Django
FastAPI
Flask
REST API Design
JWT / Clerk
Rate Limiting & Quotas
Idempotent Quota Settlement
Socket.io
WebRTC
Webhooks
REST APIs

02 / Applied AI & Pipelines

Context, provider workflows, and model output.

Google AI / GenAI
OpenAI API
Google ADK
LangChain
LangGraph
Agentic AI
RAG
Embeddings
Prompt & Context Orchestration
Usage Accounting
Multimodal Media Pipelines
TensorFlow
Keras
Pandas
NumPy

03 / Data & Persistence

Schemas, relationships, and explicit state.

PostgreSQL
Prisma
Sequelize
MongoDB
Mongoose
MySQL
SQLite
SQLAlchemy
Redis
DynamoDB
Schema Design
Loan Lifecycle Models

04 / Frontend Architecture

Accessible interaction and application state.

React
Next.js
TypeScript
JavaScript
HTML5
CSS3
Tailwind CSS
WAI-ARIA Tabs & Dialogs
State Management
Performance Optimization

05 / Delivery & Developer Tooling

Reproducible environments and reviewable changes.

Docker
Docker Compose
AWS
GCP
Linux
CI/CD
Cloudinary
Git
GitHub
Postman
Cursor
Codex
Claude Code

How I approach the build

Define the contract

Start with the product flow, API inputs, authorization boundary, and data ownership.

UGC · Clerk-scoped project endpoints

Keep state explicit

Model the states a request can enter, including provider failures, retries, and incomplete work.

Banking · pending / approved / rejected

Bound external work

Validate before a provider call. Limit usage and keep decisions that matter in application code.

Banking harness · policy + confirmation

Verify the delivery

Check the API, generated assets, source references, and deployment response before calling work complete.

Portfolio · pre-render + HTTP checks

05 / Contact

Let’s build something useful.

Good things start
with a conversation.

Have a product to build or a team to grow? I’m open to full-time, contract, and freelance opportunities in full-stack and AI engineering.

harsh.gavand.tech@gmail.com

GitHub LinkedIn

Mumbai, India · Working across the full stack

Also reachable at harshgavand2@gmail.com

A little about your idea.

Prepare an email draft. You review and send it from your email app.

I build the interface.
And the system
behind it.

Four systems. The engineering behind them.

UGC Content Studio — AI Media Generation Pipeline

Architecture & engineering decisions

Workspace: Brand Shoot #14

LendFlow — Multi-Role Lending & Underwriting Platform

Architecture & engineering decisions

DevGraph — Repository Discovery & Issue Indexing Platform

Architecture & engineering decisions

AI News Automation — Ingestion-to-Email Pipeline

Architecture & engineering decisions

One digest. Two formats.

Experience that connects the layers.

AI-Native Full-Stack Developer

Full-Stack Developer

Curious about the whole system.

B.Tech · Information Technology

Architecture before abstraction.

Contracts, identity, and request admission.

Context, provider workflows, and model output.

Schemas, relationships, and explicit state.

Accessible interaction and application state.

Reproducible environments and reviewable changes.

How I approach the build

Define the contract

Keep state explicit

Bound external work

Verify the delivery

Let’s build something useful.

Good things start
with a conversation.

A little about your idea.

Ask Harsh’s AI

What would you like
to know?

I build the interface.And the systembehind it.

Four systems. The engineering behind them.

Architecture & engineering decisions

Workspace: Brand Shoot #14

Architecture & engineering decisions

Architecture & engineering decisions

Architecture & engineering decisions

One digest. Two formats.

Experience that connects the layers.

AI-Native Full-Stack Developer

Full-Stack Developer

Curious about the whole system.

B.Tech · Information Technology

Architecture before abstraction.

Contracts, identity, and request admission.

Context, provider workflows, and model output.

Schemas, relationships, and explicit state.

Accessible interaction and application state.

Reproducible environments and reviewable changes.

How I approach the build

Define the contract

Keep state explicit

Bound external work

Verify the delivery

Let’s build something useful.

Good things startwith a conversation.

A little about your idea.

I build the interface.
And the system
behind it.

Good things start
with a conversation.