Why we built Bijirin
A free, ad-free way to tell whether a price is a good one where you live, and an honest look at how food prices are made.
Is this a good price?
When buying groceries, we often wonder what the going price is, or whether the price in front of us is a good one. Is it per kilo or per pack? Is it cheap for this state? The answer depends on where you are: the same kilo of chicken costs differently in Kuching and in Kuala Lumpur, and between a pasar and a hypermarket down the road. We know a "good" price is subjective, so Bijirin shows the range and lets you judge.
The government already publishes shop prices, but the tools to read them were cluttered with ads, hard to use on a phone, or needed an app to be installed. Bijirin runs in the browser and puts a price next to what most shops in your state charge, with no ads and no tracking.
From farm to your basket
We want to see every step of a price, from what the farmer is paid to what you pay at the till. Today we have the last steps for every item that KPDN publishes through PriceCatcher. Farm-gate prices are patchy: for a few crops we hear from friends who farm in one area, which we use only for our own reference. The middle is the hardest part, because it carries transport, storage, distance and supply and demand. It is the next thing we want to measure, once reliable data exists.
- Farm-gate What the farmer is paid A few crops, from farmers we know. For our own reference, not published.
- Collector Buys from farms, sorts and packs Not measured yet
- Wholesaler Sells on to shops and markets Not measured yet
- Retailer Shops, markets, hypermarkets KPDN PriceCatcher, every day
- You The price you pay The number Bijirin shows
Simple first, deeper if you want it
Most people want one number and a quick answer, so we keep the main pages as short as we can. The grid view gives a quick picture of each item, while the list view puts more figures side by side for comparing. If you like to dig, the what-if and drivers pages show the engine underneath.
A machine-learning model learns from about 300 items at once: seasons, festivals such as Hari Raya and Chinese New Year, and how prices drift back to their usual level. It gives a range for each of the next 1 to 8 weeks, low, middle and high, and is retrained every week.
Think of it like a weather forecast: a likely range, not a promise. Every range is scored later against real prices, and items with a weak record are pulled back towards "no change".
For when you don't know how supply or demand will move, and want to test an idea. Plain economics rather than a black box: a shock either raises the cost of bringing food to the stall (diesel, a weaker ringgit), changes how much arrives (an import cut, a flood) or changes how much people want (a festive season). The sizes of these effects are measured from real events, such as the June 2024 diesel subsidy change.
Price-controlled items are flagged, and are never shown above the government ceiling.
An assistant you can check
The assistant answers in BM, English and 中文 using the same numbers as the pages. It chooses which tools to call and writes the words, but never does the price arithmetic. Personal details are masked before anything leaves our servers, you agree before your first question, and it declines investment advice and accusations against named sellers.
- 1 Your questionIn BM, English or 中文
- 2 Personal details maskedNames, IC, phone, email, address
- 3 AI picks toolsFrom a fixed set
- 4 Tools fetch the numbersThe same data as the pages
- 5 AI writes the answerWords only, no arithmetic
- 6 Numbers checkEvery figure must match a tool result
If a number does not match, the answer is rewritten once, then replaced with a plain table built straight from the data.
Free, on a small budget
Bijirin is free for everyone, so every running cost is a choice. Prices, forecasts and what-ifs need no account. The assistant costs money for each answer, so it needs a sign-in and has a monthly allowance. We use a compact, low-cost AI model, which matched a larger one in our tests and was three times faster. We also dropped a separate analytics database (OLAP): one database holds everything, which is one thing to secure, back up and pay for. We would only consider charging if there is demand beyond the assistant's allowance, or for business needs and integrations.
Built with each part kept apart
Each part of Bijirin can do only what its job needs. The website, the API and the assistant run separately, each with its own access, and the assistant can only read data we have approved.
We have built on AWS, private servers (VPS) and Cloudflare Workers. For Bijirin we chose Google Cloud after weighing several factors: credits and running cost, the libraries and managed services we need, how the servers are set up, and how well it fits our development process.
Built to be reused
Forecasting, machine learning and actuarial methods are only worth something when they are put to work. Bijirin is where we prove them on public data. The same approach, ranges instead of single numbers, what-if questions and an honest record of past forecasts, can help companies of any size in other fields.
What we will not do
- No ads and no tracking cookies.
- No selling your data.
- No promises of accuracy we cannot show.
- No hidden sources: every one is on Sources and licences.
Want to work with us, found a problem, or have a question? Contact us.