Pharmacies
Retail chemists and pharmacy chains.
The day-to-day
- Thousands of SKUs, with substitutes for most of them
- Batches and expiry dates on every strip
- Demand that shifts with the season and the neighbourhood
Pharma software · AI research · Model training
Inwebbo builds software for pharmacies, pharma wholesalers and distributors, and the supply chain that connects them. Underneath it are AI models we research and train ourselves.
What we do
Software for pharmacies, wholesalers, distributors and the supply chain between them, built around how the trade actually works.
Our productsWe study how models handle the messy data of the pharma trade: catalogues, searches, orders, sales and supply signals.
Our researchWe curate data, then pretrain, adapt, align and evaluate our own models end to end, and ship them inside our products.
How we trainWhat our products run into tells us what to research next, and the loop starts again.
01 Products
Medicines pass through many hands before they reach a patient. We build software for the businesses along that chain: pharmacies, wholesalers and distributors, and the supply chain that links them to manufacturers.
Retail chemists and pharmacy chains.
The day-to-day
Stockists, distributors and C&F agents.
The day-to-day
From manufacturer to pharmacy shelf.
The day-to-day
Our products carry the models we research and train: they spot shifts in demand early, understand what people search for, and flag what needs attention.
Demand, search and sales signals monitored continuously, with emerging trends flagged the moment they separate from the noise.
Every search mapped to structured intent, so you see what customers want, not just what they typed.
The views that matter to each team, from the pharmacy counter to the distribution desk.
Instant notice when demand spikes, stock nears expiry or a trend turns.
A bird's-eye view across outlets, regions and channels, so growth and gaps are obvious at a glance.
Connect the systems you already run, and push insights out through an API and webhooks.
02 Research
General-purpose models stumble on the long tail of the pharma trade: brand and generic names, strengths and pack sizes, mixed-language searches, sparse and seasonal demand. That long tail is where we work.
Compact models that know catalogues, categories, generics and brands, and stay accurate on the long tail where general models guess.
Turning short, misspelt, mixed-language searches into structured intent: the product, its attributes, the urgency and the place.
Time-series models for sparse, seasonal, promotion-driven data that flag a shift early, and say how sure they are.
Answers tied to your catalogue, your documents and your numbers, with sources attached rather than confident guesses.
Domain benchmarks, calibration and failure analysis, built before the model rather than after it.
Distillation and quantization so models run fast and affordably inside the software people use every day.
Open questions
Our research agenda comes from the problems our products run into. A few we're spending time on:
03 Training
We run the whole training pipeline in-house, for the models inside our products.
Most of a model's quality is decided before training starts. We source, de-duplicate, filter and label data, and strip personal information before anything reaches a training run.
Continued pretraining on curated domain text gives a model the vocabulary of the pharma trade: brands and generics, molecules and strengths, pack sizes, regions and seasons.
Supervised fine-tuning and preference optimization turn a capable model into a dependable one: structured outputs, tool use, and knowing when to hold back.
Every model passes task-specific evaluations, robustness checks and red-teaming before it ships, and every new version has to beat the one it replaces.
We distill and quantize models so they are fast and affordable to run inside our products, release them in stages, and keep watching for drift once real traffic arrives.
How we work
If we can't measure it, we don't claim it. Evaluations come first, and they come with the model.
Pharmacies and distributors run on thin margins and little spare time. Software has to fit the work, not add to it.
Latency, cost and reliability are research problems too. We design for them from the first experiment.
Privacy-preserving pipelines, careful evaluation, and people in the loop where decisions matter.