Logging Indian meals the way you say them: roti, dal chawal and Hinglish

"2 roti, dal, thodi sabzi." That is how a lot of people describe lunch. It is also exactly the kind of meal that most food trackers handle badly.

Many apps were designed around barcodes and restaurant menus. Search for "dal" and you get a long list of near-identical entries with different numbers. Search for "sabzi" and you get nothing useful at all, because sabzi is not one dish. By the third day of scrolling through lists and weighing katoris, most people give up.

This article is about a different approach: describe the meal the way you would say it, check a draft, and confirm it. It also covers what an AI estimate of a home-cooked meal can do well, and what it cannot.

Why home-cooked Indian food is hard to log

The difficulty is not the food. It is how much a single name can cover.

• One name, many recipes. Dal can mean toor, moong, masoor or chana, thin or thick, with or without a tadka. Two households can make the "same" sabzi in completely different ways.
• Portions are described, not weighed. People think in rotis, katoris, plates and "thoda sa", not in grams. A katori in one kitchen is not the same size as in another.
• Fat is mostly invisible. Ghee on a roti, oil in a tadka or the cream in a restaurant curry can change a meal's energy a lot, and none of it is visible in a description.
• Meals are shared. When everyone takes from the same dishes, knowing exactly what was on your plate is guesswork, even for you.
• We mix languages. "Aaj lunch mein rajma chawal tha" is a perfectly clear sentence to a person. Many apps cannot handle it at all.

None of this means logging is pointless. It means the useful goal is a reasonable estimate that is quick enough to keep doing, not a precise measurement that you stop doing after a week.

Describe it, pick a portion, confirm

A simple loop works better than a database search:

1. Say what you ate in your own words. "2 roti, dal chawal" is enough. So is "poha with peanuts" or "masala dosa, sambar, chutney".
2. Pick a portion. Instead of typing grams, tap a portion size that matches what you had.
3. Check the draft. An estimate of calories, protein, carbohydrates and fat comes back as a draft. Adjust it if something looks off, such as an extra roti or a meal cooked with less oil.
4. Confirm. Only then does the meal join your log.

The confirmation step matters. An estimate you have looked at is more useful than one saved automatically, and it keeps you in charge of your own records.

What AI estimates are good at

A language model is useful here because it understands descriptions the way a person does. It can:

• Understand everyday names and mixed languages, including Hinglish, regional dish names and informal descriptions.
• Split a meal into parts. "Dal chawal with a papad" becomes separate items with their own portions.
• Give a sensible starting point based on typical home recipes, which you can then adjust.
• Keep the effort low, so you actually log meals on busy days, which is when patterns are most useful to see.

Over a week or a month, those estimates are good for seeing patterns. Are most of your meals light on protein? Do weekends look very different from weekdays? Does lunch get skipped on long work days? Those are the questions food logging answers well.

What AI estimates cannot do

It is just as important to be clear about the limits.

• It cannot see the oil, ghee or sugar. Unless you mention them, the estimate assumes a typical amount, and your kitchen may be quite different.
• It does not know your recipe or your portion. It knows common versions of a dish, not the version your family makes.
• Photos help with what, not how much. A picture can show which foods are on the plate. It cannot reliably tell how much oil went into them or how deep the bowl is.
• Micronutrients are rough estimates. Values such as fiber, sugar, saturated fat, sodium, iron, calcium, potassium and vitamin C depend heavily on ingredients, preparation and even the soil food was grown in. An estimate from a short description is at best a rough indication, never a measurement.
• It is not dietary advice for a medical condition. If you manage diabetes, kidney disease, a heart condition, pregnancy or anything similar, your nutrition needs guidance from a doctor or registered dietitian.

The honest way to use AI estimates is as a quick, consistent approximation. Consistency is what makes trends meaningful, even when individual numbers are imperfect.

Tips for better estimates

• Mention what changes the numbers. "Roti with ghee", "less oil", "restaurant paneer" and "2 katori dal" all make the estimate closer.
• Log soon after eating. Memory of portions fades quickly, especially for snacks.
• Do not chase perfection. A meal logged roughly is more useful than a meal not logged at all.
• Look at weeks, not single meals. Patterns are more reliable than any one entry.

In Velqiri

In Velqiri you log a meal the way you say it. Type "2 roti, dal chawal", tap a portion, and the coach drafts calories, protein, carbs and fat, plus optional estimates of fiber, sugar, saturated fat, sodium, iron, calcium, potassium and vitamin C. Estimates are labelled, and nothing is saved until you confirm. If you edit a meal, its micronutrient estimates are removed, because they cannot be edited. Macro nudges follow your goal.

You can also send a meal photo. An AI model lists the foods it can see so you can review a draft; the photo is processed and then discarded, not stored. You can chat with the coach in the language you write in, Hinglish included. If you log meals and nothing is logged by early afternoon, you may get one gentle midday check, which you can silence like any other moment.

The Support page explains meal drafts and lab reports in more detail, and the Privacy Policy lists what is sent to the AI providers and what is kept. Velqiri is a wellness app, available on the App Store for iPhone and Apple Watch, and the coach is an AI, so check anything important.

The short version

• Describe meals in your own words, choose a portion, check the draft, then confirm.
• AI is good at understanding everyday and mixed-language descriptions, and at giving a quick starting point.
• It cannot see oil, ghee, sugar or your exact portion, so mention them when they matter.
• Micronutrient values are rough estimates, never measurements.
• Consistent rough logging reveals patterns; for medical diets, talk to a professional.

Velqiri is a wellness app, not a medical device. Nutrition values are AI estimates, not measurements, and this article is general information, not dietary or medical advice. For nutrition guidance related to a health condition, talk to a doctor or registered dietitian.