Machine learning development in Austin
Austin has been a semiconductor town for decades, and the fabs and equipment suppliers around it produce more data than most software companies: tool telemetry, inspection images, test results for every wafer. Their people are process, yield and equipment engineers, who are hard to hire and are not hired to build production machine learning. So the models that exist often live in one engineer's notebook, run by hand when a yield excursion needs explaining.
We build the parts that turn a notebook into a system. Unlabelled defect data can be grouped with embeddings and clustering before anyone labels it; we used the same approach on work-order text to find repeat faults for FM360. Tool telemetry can be watched for drift against its own history. Anything that affects a disposition decision stays a recommendation to an engineer, logged with its inputs, rather than an automatic action. We bring the data and software expertise, your engineers bring the fab knowledge, and on the call we will be specific about how the two fit together.
The city's centre of gravity is enterprise SaaS and consumer fintech, and both bring the same backend problem early: multi-tenancy that was bolted on after the first enterprise deal, and a permissions model that started as a role column. Then the large customer asks for custom roles and SSO, the board asks for usage-based pricing, and the schema was built for neither. Unpicking that while the product keeps selling is most of what an Austin engagement looks like in the first month. On the fintech side the same pressure lands on the mobile app and the ledger: transfers that must never send twice, and balances that have to match what finance sees.
Austin has also been a hardware and semiconductor town for decades. The fabs and equipment suppliers produce tool telemetry, inspection images and test results in volumes most software companies never see, and the hardware start-ups around them reach the point where a board works on the bench and a pilot customer wants to see its data. Their engineers are process, yield and firmware specialists. The software around the hardware, from the device API and the dashboard to the pipeline that turns readings into something a customer can open, is a defined piece of work that suits an outside team.
Then there is timing. Austin has an unusual density of Series A and B product companies, and they tend to raise, hire aggressively for two quarters, then freeze. A mid-level engineer costs around $145,000 base before equity and benefits, and the platform or infrastructure hire is usually the one that waits longest. We are frequently the answer to "we have the roadmap and the budget, but the req is frozen until next quarter", because an engagement is an invoice rather than a headcount. It starts with a fixed-price two-week piece, and it can grow into an embedded engineer for as long as the roadmap needs one.
We move our working day for Texas. Four hours of every working day overlap with your morning in Central time, so stand-up, code review and the design conversation happen live with the engineer who writes the code. The code lives in your repositories from the first commit, and runbooks and architecture notes are written as we go, so whoever you hire next inherits work they can read.
We are the right fit when the roadmap is ready and the req is frozen. A defined piece of work, such as a permissions model, a device backend or a cloud setup brought under control, starts with a fixed-price two-week piece. Senior capacity arrives within days, and agencies can put the work under their own 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 Austin teams.
How do you work with teams in Austin?+
We are in Bengaluru and move our working day for Texas, so four hours of every working day overlap with your morning in Central time. Stand-ups, code reviews and design calls happen live in that window. We work in your tools, whether that is Slack, GitHub or Linear, and the person on every call is the engineer who writes the code.
Why do Austin product companies come to you?+
Austin has an unusual density of Series A and B product companies with a strong front-end team and no infrastructure hire yet. That is precisely the seat we fill: the multi-tenant backend, the permissions model, the cloud setup and the deploys that nobody senior owns. It is why more of our US enquiries come from Texas than from the coasts. We build it in your codebase, alongside the team you already have.
How does your rate compare to hiring in Austin?+
A mid-level software engineer in Austin earns around $145,000 base before equity and benefits. Our published rate is $35/hour, or $5,400 a month for an embedded engineer, with a $5,000 minimum. There is no recruiting time and no employment overhead, and work starts within days of the call. The first piece is a fixed-price two-week engagement at $2,800, so you judge us on what ships.
Our req is frozen until next quarter. Can you cover the gap?+
Yes. It is a common Austin situation, and a good fit for an engagement, because an invoice is not a headcount. The roadmap keeps moving while the req waits. Start with a fixed-price two-week piece, or embed an engineer at $5,400 a month. Runbooks and architecture notes are written as we go, so when your hire does start, they inherit work they can read and extend.
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.