Machine learning development in Delhi NCR
Anyone who has delivered parcels in the NCR knows the address problem. The same building is written a dozen ways, landmarks stand in for street numbers, and a pin code can be wrong without anyone noticing until the rider is lost. Logistics and delivery companies here lose money on this every day in failed deliveries, wrong promised dates and riders sent to the wrong sector.
Machine learning is useful here in modest, practical forms. Classifying and normalising addresses so duplicates are recognised, estimating delivery times from your own history rather than a fixed rule, forecasting demand by area so riders and stock are in the right place. We build these into the operations systems your team already uses, with monitoring that shows when they drift, and we will tell you when a set of rules does the job just as well.
Delhi NCR is not one market but three, joined by expressways and a metro. Gurugram is where much of India's consumer internet grew up: food and grocery delivery, travel, logistics, home services and fintech, alongside the Indian offices of many multinationals. Noida has a different mix, with large IT services firms, edtech, media and healthcare technology companies selling to hospitals in the US. Delhi itself holds the government, and with it every company whose customer is a ministry, a state department or a public sector undertaking.
The consumer companies share an operational problem. Their software coordinates people in the physical world, riders, couriers, technicians and sellers, on cheap phones and busy roads, and the backend has to cope with status updates that arrive late, twice or out of order. Many of them are now past the stage of growth at any cost and into the stage of unit economics, which turns engineering questions into money questions: what does each order cost to serve, and which part of the cloud bill is paying for nothing.
Companies that sell to government need software shaped like the organisation it serves. Access follows the organisation chart: a district sees its own records, a state sees its districts, a ministry sees all of them, and every approval has a proposer and an approver. Noida's healthcare technology companies need something similar for hospital customers abroad, with each customer's data kept firmly apart. Across all three cities the hiring market is crowded. The consumer companies, the multinationals and the IT services firms compete for the same senior engineers, so a defined piece of work often waits months for someone to own it.
We are in Bengaluru and share your time zone, so the working day is the same on both sides, and a problem raised in the morning gets a response that morning. You talk directly to the engineer who writes the code. Most of the work happens in your repository and a shared chat channel, and the decisions, runbooks and architecture notes are written down as we go for everyone who was not on the call.
We are the right fit when an NCR team needs senior capacity quickly: a defined build, a backend that has to stop losing orders, or the months while your permanent hire works out their notice. Start with a fixed-price two-week piece and judge us on what ships. Agencies and IT services firms can bring us in white-label under their name.
Four ways this arrives.
Someone got good results on a laptop and nobody can reproduce them. We turn it into a training pipeline with pinned data and versioned artefacts, put a serving endpoint in front of it, and record what it was evaluated on so the next version can be compared.
Hundreds of conditions routing tickets or flagging risky orders, each added after an incident. We measure what the rules get wrong on your labelled history, then test whether a classifier beats them. Sometimes the answer is fewer, better rules, and we say so.
Demand, load or staffing estimates that miss often enough to be ignored. We start with a plain seasonal baseline, backtest against the periods you actually plan around, and add complexity only where it beats that baseline on those periods.
Duplicate records, recurring faults, near-identical products under different names. Embeddings group text by meaning where keyword matching cannot, and we build the screen that lets a person correct a group rather than trust it blindly.
Repeat-fault detection that began as UMAP and HDBSCAN over 33,000 work orders and shipped as nightly incremental clustering on the same embedding model.
Read the write-up →Asked by Delhi NCR teams.
How do you work with teams in Delhi NCR?+
We are in Bengaluru and share your time zone, so calls, reviews and releases happen inside your normal working day, whether you are in Gurugram, Noida or Delhi. We work in your tools, Slack, GitHub and Linear or whatever your team already uses. You talk directly to the engineer writing the code, from the first 30-minute call onwards, and the decisions, runbooks and architecture notes are written down in your repository as we go.
Do you build for delivery and logistics companies?+
Yes. We build the systems that coordinate riders, couriers and technicians: order and shipment state that stays correct when events arrive late, twice or out of order, partner apps that keep working offline and upload proof of delivery when the network returns, and reconciliation that catches what still slips through. We also build the data pipelines that finally give everyone one answer to what an order costs to serve.
How does your rate compare to hiring in Delhi NCR?+
PayScale puts a mid-career software engineer in Gurugram at about ₹16 lakh a year. Our published rate is $35 an hour, with a $5,000 minimum. That buys a senior engineer who writes the code, works in your Slack, GitHub and Linear, and starts within days, with no recruiting round, notice period or employment overhead. Begin with a fixed-price two-week piece at $2,800 and judge us on what ships.
Can you take over a platform another team built?+
Yes, and it is a common way in. We start by reading the code and writing down what it actually does: the services, the background jobs, the parts only one person understood. The fixed-price two-week piece is usually the failure that costs you most, fixed and documented. At the end you have working code, a map of the system and a written plan for what comes next, all in your repository.
Do we need machine learning, or is a rules engine enough?+
Often a rules engine is enough, and it is cheaper to run, explain and audit. If the people who make the decision today can write it down, write it down. A model earns its place when the rules keep multiplying, the inputs are messy text or images, or the pattern shifts faster than anyone can edit the rules. We check against your history before recommending either.
Why not just call a hosted model API?+
Sometimes you should. With low volume, messy language and no labelled data, a hosted model and a good prompt ship quickly. A trained model wins when per-call pricing starts to hurt at your volume, when latency matters, when data cannot leave your systems, or when you have labelled examples a small model can learn from. We price both paths before you commit.
How much data do we need?+
For classification, usually fewer labelled examples than people fear, and embeddings can group text with no labels at all. Forecasting is the opposite: you need enough history to cover the cycles you plan around, and a single year says little about a yearly pattern. We give you an answer after looking at your data, not before.
What happens when the model gets worse?+
It will, because the world it learned from changes. We log inputs and predictions, compare them with outcomes as they arrive, and alert when accuracy or the input mix drifts past a threshold you agreed. Retraining is a scripted pipeline with an evaluation gate, so a new model replaces the old one only when it does better on the same test set.
Can you build a recommendation system?+
Yes, and we would start without a learned model. Co-occurrence and embedding similarity, served from Postgres or a vector index, beat a most-popular list for most catalogues and are easy to debug. A learned ranker comes after that, once you have enough interaction data to train and evaluate one. A learned ranker is something we build from our work with embeddings and clustering, starting from a simple baseline you can measure it against.