Generic inventory software assumes a clean purchase-to-sale path. Small manufacturers also use job-work issue and receipt documents, delivery challans that move goods without a sale, partial returns, process loss, and conversion costs. If several documents can quietly change the same stock, staff enter things twice and stop trusting the inventory figure.
02 / Vertical AI · Multi-tenant B2B SaaS · Operations
AI-Enabled Manufacturing Operations ERP
Brand: ORENX
Inventory and billing software for small Indian manufacturers, built around the material movements that generic invoice-first tools often miss: outside processing, partial returns, scrap, production, and warehouse transfers.
In each workflow, exactly one document is allowed to change stock. Quotes, orders, invoices, and planning records point back to that movement instead of changing inventory again. This rule keeps the stock figure traceable and prevents double counting.
ORENX is live at orenx.in and used daily by manufacturing businesses. It has 22 document types covering sales, purchases, inventory, production, outsourced job work, returns, and transfers, plus separate company data, GST-ready records, E-Way Bill file export, reporting, OCR, an 18-tool AI assistant, and a WhatsApp assistant linked to each user's account.
Businesses use the product daily, and it earns early subscription revenue. The next step is to measure task completion time, correction rates, and how often entries arrive through WhatsApp instead of the web app.
01
User and problem
Who needed this product, and why.
Owners and staff at small Indian manufacturers and traders who need GST-ready paperwork without losing track of material held by a subcontractor, moving between warehouses, or changing form during production.
Small factories do not run on invoices alone. Material leaves for a subcontractor, comes back as finished goods and scrap, moves between warehouses, or is consumed in production, and none of those events is necessarily a sale. I rebuilt ORENX around 22 document types, each with an explicit rule for whether it changes stock, records a future commitment, or bills work that already happened.
02
Important decisions
Decision 01
Give job work its own documents
- Why it mattered
- Materials can leave for outside processing, return as finished goods and scrap, or move between locations without a sale taking place.
- What I did
- Use separate job-work issue and receipt documents with their own stock and costing rules instead of making one delivery document mean different things in different screens.
- What it required
- Users have to learn a few more document types, but each document keeps one clear meaning and stock does not change through hidden side effects.
- The result
- Every movement records what happened and why, so inventory, supplier balances, returns, production costs, and GST records can be reconciled without guesswork.
Decision 02
Avoid a costly compliance API
- Why it mattered
- An E-Way Bill is required for many goods movements in India. Entering one manually took about 2 to 3 minutes across more than 15 fields, while the available API cost more each year than the client paid for the software.
- What I did
- Generate a GSTN-compatible JSON file from existing ORENX records for bulk upload to the government portal instead of buying the API connection.
- What it required
- The user still completes one upload step, but avoids a recurring vendor bill and the maintenance of another backend integration.
- The result
- The export runs in the browser and writes the government's required file format from data already in ORENX, without a new backend service or database migration.
Decision 03
Bring the ERP into WhatsApp, with a confirmation step
- Why it mattered
- Owners and staff of small factories already run their day on WhatsApp. Opening a web ERP to record one payment or check stock was often the step that got skipped.
- What I did
- Link each user's WhatsApp number to their ORENX account with a one-time code, then let them type requests in plain language or send voice notes. Gemini transcribes the voice note, an AI agent with 18 business tools prepares the entry, and nothing is saved until the user taps Yes.
- What it required
- Every change needs one extra tap, and invoice photos produce line items for review instead of posting directly. That is slower than full automation, but a wrong amount never reaches the books unseen.
- The result
- Users can check stock, create documents, record payments, and get ledger PDFs from the app they already use, in English, Hindi, Gujarati, Tamil, or Hinglish.
Decision 04
Show a smaller true number
- Why it mattered
- One dashboard label described a document count as material still held by vendors, and a separate production-history query silently returned no records.
- What I did
- Fix the broken query and rename each statistic to describe exactly what it counts instead of keeping a broader, more impressive label.
- What it required
- The dashboard presents a narrower number, but users know what it means and can act on it safely.
- The result
- The correction restored production history and clarified the metric without changing any of the seven rules that protect stock movement and costing.
03
What I built
What the product includes and how it works.
- 22 business and manufacturing document types, each with defined stock and costing behavior
- Dedicated flows for job-work issue and receipt, production, warehouse transfers, returns, and scrap
- GST-ready documents plus an E-Way Bill file generated in the browser for bulk government-portal upload
- A navigation and naming redesign after users struggled to distinguish similar manufacturing screens
- AI features redesigned to use fewer model calls per task, reducing wait time and per-user cost
- An AI assistant on native Gemini function calling with 18 tools for stock, documents, payments, ledgers, and reports
- A WhatsApp assistant linked to each account by a one-time code, with typed and voice requests and Yes/No confirmation before any change
- Invoice and handwritten-sheet OCR that turns photos into line items for review
How it works. ORENX keeps each company's data separate and gives every document type its own fields, status changes, and stock rules. Related inventory updates are saved together so a partly completed entry cannot leave stock inconsistent. Compliance files are generated from the same business records, while OCR and AI assistance reduce repeated data entry and help users find information.
04
What exists today
The product supports 22 types of business and manufacturing documents
These cover sales, purchases, inventory, production, job work, returns, and transfers.A third-party E-Way Bill connection would have cost about ₹11,800 or more each year
I chose a compatible file export to avoid that recurring cost. Vendor pricing can change.Entering one E-Way Bill manually took about 2 to 3 minutes
This recalled working estimate was used to decide which part of the process to automate.Manufacturing businesses use ORENX daily
Each business runs as a separate tenant with its own isolated data; the product earns early subscription revenue.Rebuilding the assistant on native function calling cut response time by about 60% and tokens per request by about 65%
Measured in development comparisons against the earlier LangChain version, not from production monitoring.$17.4K in startup credits: $14.4K from AWS for Kiro and $3K from Google Cloud for hosting
The AWS credits pay for the Kiro agentic IDE used to build ORENX, and the Google Cloud credits cover hosting.Time saved per task still needs to be measured
Usage is known, but task timing and correction rates are not yet tracked in a reliable dashboard.05
Reflection
What I learned, and what I would test next.
- A product built for one industry can handle the messy exceptions that general software treats as rare edge cases.
- Partial automation can be the better product when customers will pay for it and it removes the most frustrating step.
- A dashboard should say exactly what each number measures, even when the accurate label sounds less impressive.
Let’s discuss the product problem.