DEV Community: Vedang Jaiswal

typegone

Vedang Jaiswal — Fri, 30 Oct 2026 19:43:02 +0000

🎹 TypeTone Piano Studio X
A futuristic browser-based piano studio built using pure HTML, CSS, and JavaScript.

TypeTone transforms your keyboard into a playable multi-octave piano with real-time synthesized audio, chord support, recording, looping, custom sound engines, and an interactive piano UI.

✨ Features
🎹 Fully playable keyboard piano
🎵 3 octave key mapping
🎼 Chord system using _
🔊 10 synthesized sound engines
🌌 Dream pad synth
🔔 FM bell synthesis
🎛 Additive organ synthesis
🔥 Warm analog-style tone
🎸 Bass low-pass synth
⚡ Bright electric synth
⏺ Record & save melodies
🔁 Loop playback
🎚 Tempo / sustain / volume controls
🌗 Light & dark theme
📱 Mobile responsive UI
🎨 Modern futuristic interface
⌨ Key Mapping
Octave 1
Q W E R T Y U → C D E F G A B

1 2 3 4 5 → C# D# F# G# A#

Octave 2
Z X C V B N M → C D E F G A B

A S D F G → C# D# F# G# A#

Octave 3
I O P H J K L → C D E F G A B

6 7 8 9 0 → C# D# F# G# A#

🎼 Chord System
Use _ between keys to play chords.

Example:

q_e_t

Plays: C + E + G

Minor chord example:

q_3_t

Plays: C + Eb + G

📝 Song Format
Spaces = pause
_ = chord
/ = ultra-fast separator
New line = longer pause
Example melody:

3_5(3spaces)4(2spaces)2(4spaces)3(4spaces)3_5(3spaces)4(2spaces)2(4spaces)1

🎛 Sound Engines
Sound Description
Classic Pure sine piano
Soft Triangle wave
Retro Square wave
Electric Sawtooth synth
Warm Filtered analog tone
Dream Detuned ambient pad
Organ Additive harmonic organ
Bell FM metallic bell
Bass Deep filtered bass
Bright High-shelf electric tone
🚀 How To Run
Simply open the HTML file in any modern browser.

No installation required.

Works best in:

Google Chrome
Microsoft Edge
Firefox
⚠ Browser Audio Note
Some browsers block audio until user interaction.

Click anywhere on the page once before playing if audio does not start immediately.

🛠 Built With
HTML5
CSS3
Vanilla JavaScript
Web Audio API
🎹 Created By
TypeTone Piano Studio X Built for typists, musicians, and synth lovers.

TypeTone Piano Studio X

Vedang Jaiswal — Wed, 21 Oct 2026 20:59:39 +0000

https://veduisback.github.io/piano_for_typists/
🎹 TypeTone Piano Studio X
A futuristic browser-based piano studio built using pure HTML, CSS, and JavaScript.

TypeTone transforms your keyboard into a playable multi-octave piano with real-time synthesized audio, chord support, recording, looping, custom sound engines, and an interactive piano UI.

1 2 3 4 5 → C# D# F# G# A#

Octave 2
Z X C V B N M → C D E F G A B

A S D F G → C# D# F# G# A#

Octave 3
I O P H J K L → C D E F G A B

6 7 8 9 0 → C# D# F# G# A#

🎼 Chord System
Use _ between keys to play chords.

Example:

q_e_t

Plays: C + E + G

Minor chord example:

q_3_t

Plays: C + Eb + G

📝 Song Format
Spaces = pause
_ = chord
/ = ultra-fast separator
New line = longer pause
Example melody:

3_5(3spaces)4(2spaces)2(4spaces)3(4spaces)3_5(3spaces)4(2spaces)2(4spaces)1

No installation required.

Works best in:

Google Chrome
Microsoft Edge
Firefox
⚠ Browser Audio Note
Some browsers block audio until user interaction.

Click anywhere on the page once before playing if audio does not start immediately.

🛠 Built With
HTML5
CSS3
Vanilla JavaScript
Web Audio API
🎹 Created By
TypeTone Piano Studio X Built for typists, musicians, and synth lovers.

ASIA_X Multi-Horizon RL Research

Vedang Jaiswal — Tue, 13 Oct 2026 11:38:32 +0000

ASIA_X Multi-Horizon RL Research

repo: https://github.com/Veduisback/olymptrade-rl

AI-based research project for analyzing ASIA_X (Asia Composite Index) quote data using a multi-horizon neural network.

The project explores whether a learned model can estimate the directional outcome of different future time horizons from historical price/quote sequences.

Research / educational project. This repository is not intended to provide financial advice or guarantee profitable trading decisions.

Overview
The project uses historical ASIA_X quote data to train a neural network that evaluates multiple possible actions across several future horizons.

The model considers:

WAIT
BUY
SELL
across six prediction horizons:

5s
15s
30s
45s
60s
120s
Instead of using manually written trading rules, the model learns relationships between the input state and historical outcomes.

Architecture
The current model is a multi-head neural network.

Quote History
│
▼
State Representation
│
▼
128-unit layer
│
▼
128-unit layer
│
▼
64-unit layer
│
├────────► 5s → WAIT / BUY / SELL
├────────► 15s → WAIT / BUY / SELL
├────────► 30s → WAIT / BUY / SELL
├────────► 45s → WAIT / BUY / SELL
├────────► 60s → WAIT / BUY / SELL
└────────► 120s → WAIT / BUY / SELL
Model configuration
Property Value
State size 22
Hidden layers 128 → 128 → 64
Horizons 5, 15, 30, 45, 60, 120 sec
Actions per horizon 3
Total outputs 18
Discount factor 0.95
Learning rate 3e-4
Framework PyTorch
Important Training Note
Although the project uses the name DQN, the current V3 implementation is more accurately described as direct offline reward prediction.

The training target is the historical reward for each action/horizon combination.

The current training step effectively learns:

state → expected historical reward
rather than performing a traditional bootstrapped DQN update using:

reward + gamma × max(next_state_Q)
This distinction is intentional and documented as part of the current research stage.

Dataset
Raw quote files use the following format:

quote_index,timestamp,price,pair
0,1720000000.123,6153.12,ASIA_X
1,1720000000.641,6153.18,ASIA_X
2,1720000001.205,6153.14,ASIA_X
Dataset structure
The data is divided into numbered segments.

Training:

segments 0–20
Unseen evaluation:

segments 21–25
The unseen segments are kept separate from training so that model performance can be evaluated on data that was not used during training.

State Representation
The model uses a state vector of size:

22
The state is constructed from recent quote history and derived market information.

A lookback window is used to provide the model with short-term temporal context rather than a single isolated price.

Reward System
For each horizon, the dataset generates rewards for all three actions.

Conceptually:

WAIT → wait reward

BUY → positive reward if future price > entry price
negative reward otherwise

SELL → positive reward if future price < entry price
negative reward otherwise
Current reward values include:

WIN_REWARD = +0.85
LOSS_REWARD = -1.0
The reward structure reflects an 85% payout assumption used during the research experiments.

Training Pipeline
The general workflow is:

Raw Quote CSV
│
▼
Dataset Loader
│
▼
State Construction
│
▼
Multi-Horizon Reward Generation
│
▼
Neural Network Training
│
▼
Model Checkpoint
│
▼
Unseen Evaluation
Training uses segments 0–20.

Evaluation uses separate segments 21–25.

Repository Structure
olymptrade-rl/
│
├── rl/
│ ├── multi_horizon_agent_v3.py
│ ├── multi_horizon_dataset_v3.py
│ ├── train_multi_horizon_v3.py
│ ├── live_ai_trader_v1.py
│ └── ...
│
├── models/
│ └── rl/
│ └── multi_horizon_dqn_v3.pt
│
├── data/
│ └── ...
│
├── requirements.txt
├── .gitignore
└── README.md
Large datasets, trained checkpoints, logs and other generated files are intentionally excluded from Git where appropriate.

Installation
Clone the repository:

git clone https://github.com/Veduisback/olymptrade-rl.git
cd olymptrade-rl
Create a virtual environment:

python -m venv .venv
Activate it on Windows:

..venv\Scripts\Activate.ps1
Install dependencies:

pip install -r requirements.txt
Training
The training script can be run as a Python module:

python -m rl.train_multi_horizon_v3
Training uses the configured training segments and saves the resulting model checkpoint.

Evaluation
The model should be evaluated against data that was not used during training.

The current evaluation set is:

segments 21–25
This separation is important because evaluating on the training data can give an overly optimistic estimate of model performance.

Current Evaluation
An evaluation on unseen segments produced approximately:

Job Applicant Portfolio

Vedang Jaiswal — Thu, 08 Oct 2026 13:15:23 +0000

🚀 Built a SaaS platform that turns your resume into an AI-powered job applicant chatbot!

live project: https://veduisback.github.io/online-job-applicant-portfolio/index.html
repo: https://github.com/Veduisback/online-job-applicant-portfolio
backend: https://online-job-applicant-portfolio.onrender.com/
What if recruiters could simply chat with your portfolio instead of going through a long resume?

That’s the idea behind this project. 👨‍💻

Users can enter their professional details — skills, education, projects, experience, GitHub, LinkedIn, and more — and the platform creates a personalized AI chatbot that can answer questions about their profile.

💡 Key features:

📝 Create your professional profile
🤖 Generate a personalized AI job applicant agent
💬 Let recruiters ask questions naturally
🔗 Share your unique applicant profile
⚡ SaaS-style architecture designed to support multiple applicants
🧠 AI responses grounded in the applicant's own information

The goal is to make traditional resumes more interactive, conversational, and useful for recruiters.

Built with technologies including React, Node.js/Express, Firebase, and OpenRouter.

I'm sharing this as part of Hacktoberfest and would love to see contributors improve the project, add new features, improve the UI/UX, and experiment with better AI capabilities. 🚀

If you're interested in AI agents, SaaS, full-stack development, or open source, this is a project worth checking out!

This is a sample post!

Vedang Jaiswal — Wed, 07 Oct 2026 12:56:12 +0000

yo vedu this side, i am 2nd year student learning computer science and want to build things which pay my bills (im in debt 😃)