LSGZ logo
LSGZ·loading
Available for AI product work

Building AI products
that people actually use.

I'm Lovepreet Singh, an AI-focused developer who enjoys transforming ideas into real products using modern web technologies, APIs, and artificial intelligence.

GitHub
1000+Users served
4Products shipped
7AI experiments
0+
Users Served
Across live products & bots
0+
Monthly Active Users
On EnhanceIt alone
0
Products Shipped
NextLecture · EnhanceIt · StreamPoint · PocketDev
0
AI Experiments
In the lab & beyond
Products

Things I've built.

Products people use, experiments that taught me something, and ideas I couldn't resist trying.

I don't build to fill a portfolio. The portfolio exists because I keep building.
Live
Try it Code

NextLecture

Never miss a lecture again.

01
EducationAndroidWeb AppTimetable
Android + Web
Platforms
Live
Status
GNDEC
College

NextLecture — A Simple Timetable That Actually Reminds You

I accidentally missed a lecture in college because I got too tired in one class and literally forgot I had another. I straight went home, and then it hit me: "oh shit, I had a lecture left." So I decided to make a timetable app that can remind me before lectures. Later I added many features, and now many students in college actually use NextLecture because it provides a simple interface for the college timetable rather than searching through the college website. The product spans a native Android app and a mobile-first web companion at nextlecture.vercel.app: offline-first timetable with live next-lecture countdown, local AlarmManager reminders that work without network, hybrid HTML + Gemini parsing of the official GNDEC timetable, multi-year section discovery, attendance tracking, previous-year papers, and a PWA web dashboard that answers what is next, when it starts, and where to go.

What I built
  • Next-lecture card with live countdown and free-period detection
  • Local offline lecture reminders (AlarmManager + boot reschedule)
  • Official GNDEC timetable fetch with ETag / Last-Modified caching
  • Hybrid deterministic + Gemini AI parsing for ambiguous cells
  • Dynamic group / multi-year section discovery from department sites
  • Offline-first Room cache and full day view (completed / happening / upcoming)
  • Attendance marking with target %, max misses, and recovery guidance
  • Student profile lookup from official 2026 branch PDFs
  • Previous-year papers and PYQ RAG support
  • Mobile-first web app (PWA) at nextlecture.vercel.app
  • Announcements feed and in-app GitHub release updates
Stack
KotlinAndroidRoomAlarmManagerJsoupGeminiTypeScriptReactViteVercelSupabase
I built it so I would never miss a lecture again. Now many students in college use it because the college website was never this simple.
Try it GitHub
Live
Try it Code

EnhanceIt

Give your photos a second life.

02
AIImage ProcessingWeb App
200+
Monthly Active Users
1000+
Registered Users
Live
Status

EnhanceIt — AI Image Enhancement, Without the Complexity

Behind that simple interaction is an image-processing workflow built around AI restoration models and external inference infrastructure. I worked through deployment limitations, API behavior, image retrieval, processing states, failures, and the UI needed to make all of that feel like one seamless product. Today, EnhanceIt serves 200+ monthly active users with over 1000 registered users.

What I built
  • AI-powered photo enhancement and restoration
  • Before/after comparison experience
  • Image upload and processing workflow
  • CodeFormer-based restoration
  • Hugging Face / Gradio integration
  • Python backend
  • Error and processing-state handling
  • Responsive interface
  • Analytics and SEO
  • Public deployment on Render
Stack
PythonCodeFormerHugging FaceGradioHTMLCSSJavaScriptRender
What began as an image-processing experiment became a tool people actually use.
Try it GitHub
Live
Try it Code

StreamPoint

Your starting point for streaming.

03
WebDiscoveryEntertainment
Live
Status

StreamPoint — Finding Where to Watch Shouldn't Be the Hard Part

It doesn't host media itself. Instead, it focuses on discovery—giving users a cleaner starting point before they leave for the service they want. The project became particularly meaningful when people around me stopped calling it "your streaming website" and simply started calling it StreamPoint.

What I built
  • Streaming-platform directory
  • Movie and anime categories
  • Fast navigation between sources
  • Search/discovery-oriented interface
  • Responsive mobile experience
  • Dark cinematic visual identity
  • Custom StreamPoint branding
  • Public deployment
Stack
HTMLCSSJavaScriptCloudflare Pages
One place. Many destinations. That's StreamPoint.
Try it GitHub
Beta
View Project Code

PocketDev

Development shouldn't require a desk.

04
Developer ToolsAIAndroid
In Development
Status

PocketDev — A Development Environment That Fits in Your Pocket

I constantly found myself generating code with AI, copying it, switching to an Android code editor, pasting it, testing it, returning to the AI, and repeating the process. Android had code editors. AI coding tools existed. But I couldn't find an Android editor that combined the two in the way I wanted. So PocketDev became my attempt at building one. Rather than squeezing a desktop IDE onto a small screen, the goal is to make AI part of the editor itself—helping write, understand, debug and work with code in an interface designed around a phone.

What I built
  • Mobile-first code editing
  • Smart inline AI suggestions
  • AI code generation
  • AI-powered bug detection
  • Code explanation
  • Automatic debugging and fixing
  • AI-assisted autocomplete
  • Swipe-to-accept suggestions
  • Partial swipe for multiline completions
  • Android-focused touch interactions
  • GitHub-oriented development workflow
Stack
AndroidAI APIsJavaScriptGitHub Actions
Some of my own projects were built from a phone. PocketDev asks how much better that experience could become.
View Project GitHub
Experiments & Lab

Not everything needs to become a product.

Some things I build simply because I want to know what happens when I try. These aren't failed products — they're things I built because I wanted to understand something.

Experiments

LSGZ Personality Clone

What if an AI could sound a little more like me?

Shipped

Most AI assistants are deliberately generic. I wanted to experiment with something different: whether an AI could maintain a recognizable communication style and personality. I collected conversational data, explored base-model selection, training, epochs and loss, and eventually trained a model around patterns from my own conversations. It wasn't intended to become another general-purpose chatbot. It was an experiment in making interaction itself part of the model experience.

LLMPersonalityModel Training
Explored
  • Model training
  • Conversational datasets
  • LLM behavior
  • Persistent conversational characteristics
  • Persona construction
  • Customized response style
  • Hugging Face deployment
  • Personalized AI interaction
The experiment wasn't just about what an AI says—it was about whether how it says it could become recognizable.
View Experiment

Learnigo

An AI learning companion built around asking questions naturally.

Exploring

Learnigo grew out of my earlier chatbot experiments and became an exploration of how conversational AI could support students, particularly where access to educational resources may be limited. Instead of limiting interaction to typed questions, the concept explored multiple ways for students to communicate with an assistant. The project also became a testing ground for ideas that later influenced how I approached other AI products.

AIEducationMultimodal
Explored
  • Conversational question answering
  • Image understanding
  • Voice input
  • Multilingual interaction
  • Educational assistance
  • Telegram integration
  • Online and local LLM approaches
Sometimes a prototype is valuable not because it becomes the final product, but because of everything it teaches you to build next.

Lab

Conversational AI

Shipped

Learning how AI conversations work by building them.

A collection of Telegram-based AI experiments exploring conversation memory, multimodal input, group interactions, custom personas, voice recognition, and different LLM providers.

Telegram Bot APIPythonFlaskGroqLLM APIsMemoryVisionSpeech
Progress100%

Local LLMs

Active

How much AI can you run on a phone?

Experiments with running language models locally on Android through Termux, Ollama, llama.cpp, and GGUF models—including Qwen2.5 0.5B—to understand what useful offline AI looks like under tight hardware constraints.

Ollamallama.cppGGUFQwenTermuxOn-device AI
Progress70%

Voice AI / RVC

Active

Experiments in teaching machines a voice.

Explored Retrieval-based Voice Conversion, model training, datasets, speech processing, and AI voice pipelines—including training a custom RVC voice model.

RVCVoice ConversionModel TrainingAudio Processing
Progress55%

Generative Image Lab

Shipped

Pixels, models, APIs and lots of failed requests.

Experiments across generative-image and restoration models, Hugging Face Spaces, Stable Diffusion/SDXL workflows, CodeFormer and image-processing APIs. Some of this experimentation eventually evolved into EnhanceIt.

Stable DiffusionSDXLCodeFormerHugging FaceImage Processing
Progress100%

Terms of Service Summarizer

Archived

Making the text nobody reads easier to understand.

An experimental tool for turning lengthy Terms of Service documents into shorter, understandable summaries using AI. Archived after reliability limitations made it unsuitable for the level of accuracy I wanted.

AIText ProcessingSummarization

And this website?

That's a project too.

LSGZ.dev

The website behind the builder. My personal corner of the internet—a custom-designed home for the products I ship, experiments I try, and things I learn along the way.

This website wasn't generated from a template or created from one giant prompt. It evolved section by section through dozens of iterations. I used different AI systems throughout the process for ideation, code, debugging, critique, and refinement—then brought those pieces together into one coherent product. And much of that process happened from an Android phone.

Read My Story
A portfolio where apparently every idea became an amazing successful product feels fake. "I built it. It wasn't reliable enough. I archived it." shows judgment.
My Story

It started with a phone.

I didn't begin with a CS roadmap, a powerful dev setup, or even a clear idea of what I wanted to build.

LSGZ

I started by making small static HTML websites and hosting them from an OPPO A15 — a fairly average Android phone. I had no laptop, so whenever the phone couldn't do something, I learned to find another way.

Termux for scripts. Colab and Kaggle when I needed a GPU. GitHub Actions to compile Android APKs I couldn't build locally. Slowly, I built my own dev environment out of whatever was freely available on the internet.

That approach led to a Groq-powered Telegram chatbot, a passport-photo pipeline my dad actually used in his shop, and eventually to EnhanceIt — a free image enhancer that quietly grew to 1000+ users without a launch, ad, or subscription.

Voice-conversion experiments for fun. A fine-tuned model that talks a little more like me. StreamPoint, a streaming directory a friend called by name instead of "that website you made." And PocketDev, an AI-powered Android code editor — built on an Android phone, compiled through the same GitHub-Actions loop it was designed to escape.

I don't pretend I wrote every line of code myself. I use AI aggressively — as a coding tool, debugger, researcher, designer — but deciding what should exist, connecting the pieces, and shipping it is mine.

Built from a phoneAI as a tool, not a crutchAlways finding another way
The long version

Read my full story

Every chapter — from the first HTML page hosted on a phone to the bugs, the 5px padding loops, the logo drawn on paper, and why I think context windows are part of a dev environment. Same story, more room to breathe, with visuals along the way.

Tech Stack

Tools I reach for first.

The stack changes as the field does. These are the tools I've shipped with recently — grouped by what they're actually for.

Language01 / 06
Frontend02 / 06
Backend03 / 06
AI / ML04 / 06
Infra05 / 06
Tooling06 / 06
Open Source

Activity on GitHub.

A snapshot of recent repositories and a placeholder contribution graph. The real one loads from the GitHub API in production.

@lsgzt

Contribution activity

Last 12 months · 1,248 contributions

LessMore
1,248
Contributions
23
Public repos
12
Followers

nextlecture-android

GNDEC timetable & lecture reminder app for Android.

Kotlin00

pocket-codex

AI-powered code editor for Android with on-device assistant.

JavaScript428

lsgz-personality-clone

Fine-tuned LLM that mirrors a personal writing style.

Python275

enhanceit

Diffusion-based image enhancement web app.

Python356

streampoint

Minimal directory of free streaming websites.

HTML183
Contact

Have an idea worth building?

I'm currently open to AI product work, collaborations, and the occasional interesting side-project. Drop me a line.

Typical reply time: 1–2 days.

GitHub
@lsgzt
Email
lovepreet.singh.dev@gmail.com
LinkedIn
in/lovepreet-singh-3b3588287
Hugging Face
@lsgz
Currently available
Open to AI product work and collaborations.