JDM Pulse

JDM Pulse shows the true all-in cost of importing a Japanese car to Bangladesh, live from the auction; see the final price before you get any more surprises! v1 out soon for testing.

Comment

JDM Pulse: Japanese Car Auction Analytics

The Inspiration: From Showrooms to Smartphone

Project led by Nirjhor, Tanjim Mohamed

The traditional model of importing cars into developing nations like Bangladesh is broken. It's a financial black box, controlled by a handful of large importers who operate expensive showrooms and pass on massive overheads to the customer. For an individual buyer, the dream of owning a high-performance JDM (Japanese Domestic Market) vehicle is mired in uncertainty, hidden fees, and a complete lack of price transparency.

JDM Pulse is our answer. We envisioned a future where anyone in Dhaka, or any other city in the world, could bypass the middlemen entirely. A future without physical showrooms, where the entire process—from discovering a car at a live Tokyo auction to calculating its final landed cost down to the last Taka—happens on a single, intuitive dashboard.

Our project is the first step in building a full-fledged, tech-first, dropshipping business for vehicles. We are creating a platform that connects Japanese sellers directly to global buyers, using predictive analytics and real-time data to make cross-border vehicle trade as simple as ordering a product online. We're not just building a dashboard; we're building the engine for a new kind of automotive company.

What It Does

JDM Pulse is a predictive analytics platform that delivers unparalleled clarity and control to the JDM import process. It is designed to be the central nervous system for our future dropshipping operation.

MVP Core Features:

  1. Curated Live Auction Feed: The dashboard showcases a filtered stream of high-demand luxury, sports, and supercars from Japanese auctions, providing all critical data points in a clean, digestible format.
  2. Predictive Bid Guidance (ML): Before a user even thinks about a price, our machine learning model, trained on historical auction data, predicts the likely winning bid for any given vehicle. This provides an immediate, data-backed valuation.
  3. Instantaneous Landed Cost Calculation: The user can input their desired bid or use our prediction, and the application instantly calculates the Total Landed Cost in their local currency (BDT for our MVP), breaking down every single component of the price.
  4. Interactive Financial Visualization (Powered by Plotly): A stunning, interactive donut chart visualizes the entire cost structure—from the auction price in Japan to the complex web of Bangladeshi import duties. This transparency builds ultimate user trust.
  5. Apple/Adaline-Inspired UI/UX: The entire experience is wrapped in a sleek, minimalist interface inspired by Apple's design philosophy. We plan to use Framer Motion and selectively GSAP (GreenSock Animation Platform) for fluid, meaningful animations, ensuring the webapp feels as premium as the cars it showcases.

How We're Building It: A Guide to a Flawless MVP

This is our detailed, agentic workflow for any senior engineer or data scientist to replicate our MVP. Our frontend is a pre-existing, responsive PHP/Laravel application.

Step 1: Sourcing & Building the < 75MB Hackathon Dataset

To build a realistic demo, we need real-world data.

Data Sourcing Strategy:

  • Challenge: Premier auction houses in Japan (like USS, TAA) and data aggregators (like TCV - Trade Car View) restrict their live data APIs to licensed dealers.
  • Hackathon Solution (Web Scraping): For this MVP, we are writing a Python script using Beautiful Soup and Requests to scrape publicly available "sold" vehicle data from portals that list past auction results. This gives us the most crucial feature for our ML model: the winning_bid_jpy. We focus on high-value models (Lexus LX600, Toyota Land Cruiser 300, Porsche 911, Nissan GT-R) from the last 3 years to create a rich, focused dataset. ### Dataset Construction & Compression:
  • Our script scrapes ~15,000 vehicle records, capturing all essential fields (make, model, year, mileage_km, engine_cc, auction_grade, and critically, winning_bid_jpy).
  • The raw data is saved to a temp_auction_data.csv.
  • To meet the < 75MB limit and optimize for analytics, we convert the CSV to the Apache Parquet format. Parquet is a columnar storage format that offers superior compression and significantly faster query performance for analytics tools. Command: pandas.read_csv('temp_auction_data.csv').to_parquet('jdm_pulse_dataset.parquet', compression='gzip')
  • The final jdm_pulse_dataset.parquet file is highly compressed, easily fitting within the 75MB limit, and is primed for our Python engine.

Step 2: The Predictive & Financial Engine (Python Core)

This is the brain. A Python script (engine.py) powered by Scikit-learn and Pandas.

  1. Proxy Bidding ML Model:
  2. We use the jdm_pulse_dataset.parquet to train a regression model. The goal is to predict winning_bid_jpy.
  3. Features: year, mileage_km, engine_cc, auction_grade
  4. Model: We use a GradientBoostingRegressor from scikit-learn due to its high accuracy with tabular data.
  5. The trained model is saved as bid_predictor_model.joblib.
  6. Core Calculation Algorithm:
  7. The engine exposes a single, powerful function: predict_and_calculate(vehicle_details: dict, user_bid_jpy: int) -> json. Workflow:
  8. Load the ML model (bid_predictor_model.joblib).
  9. Use the model to generate a predicted_winning_bid_jpy for the selected vehicle.
  10. Execute the full landed cost calculation based on the user_bid_jpy. This involves applying all Japan-side fees, freight costs, and the complex, multi-layered Bangladeshi tax structure (codified in bd_tax_rules.json).
  11. Return a JSON object containing the predicted_winning_bid_jpy, and the complete cost breakdown for Plotly visualization.

Step 3: UI/UX & Frontend (Plotly Studio, Laravel, GSAP)

This is where the data meets world-class design.

  1. Rapid Prototyping in Plotly Studio: We upload a sample CSV output from our engine to Plotly Studio. We design the core "Landed Cost Breakdown" donut chart and a "Market Price vs. Mileage" scatter plot. This code-free environment allows our UI designer to perfect the data visualization aspect. We export the data and layout JSON structures from Plotly Studio, which serve as direct templates for our live implementation.
  2. Frontend Implementation (Laravel + TailwindCSS + Framer Motion / GSAP):
  3. Aesthetic: Our Laravel frontend uses Tailwind CSS to achieve the clean, modern aesthetic of Shadcn UI.
  4. Interaction: When a user selects a car, an AJAX call is made to a Laravel API endpoint.
  5. API Call: The Laravel controller executes the Python engine.py script, captures the JSON output, and returns it to the frontend.
  6. Dynamic Visualization: Using Plotly.js, we dynamically update the donut chart with the data from the API response.
  7. GSAP Animation: We use GSAP to animate the transition. As the new data arrives, the numbers on the dashboard don't just change—they elegantly count up or down, and the Plotly chart segments smoothly animate to their new positions. gsap.to(".total-cost-display", { textContent: new_cost, duration: 1, ease: "power1.inOut" });

Accomplishments That We're Proud Of

We successfully transformed a complex, real-world financial problem into a beautiful and intuitive user experience. Our proudest achievement is the seamless integration of a predictive ML model with a high-fidelity, animated frontend. The agentic workflow—where a user's click triggers a prediction, a complex calculation, and a GSAP-animated Plotly visualization in under a second—is a testament to modern web application architecture. We have built a prototype that not only works but feels like a finished, premium product.

What's Next for JDM Pulse

The Vibe-a-Thon MVP is the cornerstone of our business. Our roadmap is clear:

  1. Live API Integration: Move from a static dataset to live API feeds from Japanese auction houses, secured through partnerships with licensed exporters.
  2. Build the Dropshipping Backend: Develop the logistics and payment platform to handle the entire transaction lifecycle, from placing the final bid to final delivery in Bangladesh.
  3. Aggressive User Acquisition: Launch a targeted social media marketing campaign showcasing the platform's transparency and ease of use, directly challenging the traditional importers.
  4. Global Expansion as ML B2B, B2C Hybrid SaaS: Replicate the model by incorporating the tax and import regulations for other LDCs, truly democratizing the global JDM market.

Built With

Share this project:

Updates