Applied AI

AI that works inside the systems you already run

Start an AI fit check Talk to an engineer

We add AI where it saves a person time inside an existing workflow: a purchase order drafted from a supplier's email, a stock entry flagged because the rate looks wrong, a supplier invoice read into a draft. Each feature has a metric agreed before we build, an eval set it must pass, a cost budget and a named approver. It runs in your tenancy or on the device and never trains a third-party model. Four AI products are in build; none is live yet.

The offers

OfferWorkflowModel typeData it needsHuman checkpoint

AI in ERPNext workflowsDrafting, flagging, classifying inside existing formsLanguage models behind a gateway, with retrieval from your own recordsThe records involved, an approval ruleA named approver on every posting

Document-to-ERP extractionInvoices, GRNs, certificates of analysis, KYCDocument understanding models with field-level confidenceSample documents, the target form, the masters to match againstBelow-threshold fields are reviewed before save

Voice agents for operationsSupplier confirmations, payment reminders, appointment reminders, delivery schedulingSpeech recognition and synthesis with a scripted, bounded dialogue, Hindi firstPhone numbers, the record to update, the questions to askExceptions go to a person; every call is transcribed and stored

Vision at the edgeDock, line, yard or queue monitoring; safety zones; countingDetection and tracking models on Jetson, Hailo or a GPU serverCamera streams, a zone definition, a few hundred labelled framesA supervisor acts on events; the model raises them

Hiring and screeningCV screening, tailored proctored exams, live codingLanguage models for scoring and question generation; a proctoring layerJob descriptions, criteria, applicationsHR approves shortlists and results

How the work runs

Applied AI runs a short route of its own: a fit check, a pilot with a person at the checkpoint, then scale. The same gate language as an implementation.

Fit check
Pilot with a person at the checkpoint
Scale

YOU SIGN ATPilot go or no-goReadiness reviewHandover

How we implement, waypoint by waypoint →

In build

ProductWhat it doesStatus

Document ExtractionInvoices, GRNs and certificates read into draft documents with line-level confidence. Lines below the threshold wait for a person; the model submits nothing.In build

Vision AIDetection on the camera or on site, so footage never leaves the premises. The system raises an event; a supervisor decides.In build

Voice AssistantOutbound calls in Hindi that confirm, remind and record the answer in your system. Exceptions go to a person.In build

Hiring and screeningCVs scored against the job on stated criteria, an exam generated for each candidate, and recruiters see the reasoning, not only the score. HR approves every shortlist.In build

None of these runs at a client yet. Each will be called live only when a client runs it in production, and its case will be published with the numbers.

How we deliver and govern it

A number first
Before we build, we agree the metric: calls confirmed per day, invoices auto-posted, events caught, hours returned. We baseline it and report against it monthly.
An eval set for every feature
A held-out set of real documents, calls or frames that the feature must pass before release and after every model change.
A cost budget
Cost per call, per page or per frame is set in advance and monitored; the feature falls back to a manual queue rather than overspend.
A provider-agnostic gateway
Models are called through a gateway Bizmap runs, with budgets and attribution per client and project. A model can be swapped behind it without a rewrite.
Your tenancy or the edge
Inference runs on your cloud account, on your server or on the device. Nothing you send is used to train a third-party model.
Human sign-off points
Every feature has a named role that approves, and a documented path for what happens when the model is wrong.
Honest labels
Where we do not yet have a measurement we say "expected" and show the method.

Where AI fits in our other services

Service: Where AI fits
ERP and business platforms: Supplier dispatch confirmation · Purchase invoice capture · Stock entry anomalies
Custom product development: Document intake · Support ticket triage · Report narrative
Mobile apps: On-device document capture · Voice to form in Hindi · Photo quality check
Geospatial and mapping: Change detection · Counting from imagery · Address cleaning

Technology

Language models (hosted, via gateway)
Bizmap AI gateway (provider-agnostic)
Document understanding models
Speech recognition and synthesis (Hindi)
Detection and tracking models
NVIDIA Jetson Orin + DeepStream
Hailo-8
Sony IMX500 edge-AI camera
IP cameras (RTSP)
Central GPU server
ERPNext

FAQ

Is any of this live?

Not yet. Document Extraction, Vision AI, Voice Assistant and Hiring and screening are in build. We call a feature live only when a client runs it in production.

What is an AI fit check?

One workflow, looked at properly. We measure the baseline, check the data, score the fit and give you a build plan with the metric we would commit to.

How do you stop the model making things up?

It works from your records, not from general knowledge: retrieval from your own documents, a confidence threshold per field, and an exception queue for anything below it. Every feature ships with an eval set of real cases it must pass, and the model never posts a transaction on its own.

Where does our data go?

To your cloud account, your server or the device. Calls to hosted models go through our gateway to the provider you approve, under contracts that exclude training. Nothing is stored outside the tenancy you choose.

Does it work in Hindi and Marathi?

Hindi first for voice. Marathi and other Indian languages follow the same pipeline and are scoped per project.

What stays human?

Every approval, every posting, every decision that costs money or affects a person. The model drafts, flags, classifies and confirms; people decide.

Pages in this section

AI in ERPNext and business workflowsDrafting, flagging and classifying inside the system you already run, with a named approver.How we deliver and govern AIA number agreed before we build, an eval set, a human checkpoint, your tenancy.