SWAP Translators is a certified translation agency operating in Colombia, the United States, Ecuador and Australia. It used to quote every job by hand over WhatsApp and spreadsheets. WebMinds, an AI and software development company based in Atlanta, Georgia, serving the US and Latin America, built SwapAI: an AI quoting system for a translation agency that grew into a full operating platform. Clients upload a scan or a phone photo and get a price and delivery date in seconds, pay in three currencies, and every paid order is routed to the right translator, checked by quality control and delivered without a single coordination chat.
Key facts
- 12 systems in production sharing one database and one pricing engine across five channels.
- 1,600+ quotes generated and 4,900+ documents processed since April 2026.
- 1M+ words paid across 35 languages and 4 certification types.
- 100+ translators (in-house and freelance) assigned automatically since October 2026.
- Six days from project kickoff (March 27, 2026) to an AI quoting MVP in production (April 2, 2026).
- 3 currencies and 7 payment methods, with automatic DIAN e-invoicing in Colombia.
Context: a certified translation agency running on chats and spreadsheets
Certified translation is not a casual service. SWAP translates legal documents for visas, immigration and official procedures, so a wrong price or a wrong delivery has real consequences for the person on the other end. Like many professional services firms, its expertise was strong while the operation around it was held together by people and spreadsheets.
Before SwapAI, every quote started with a document sent over WhatsApp or email. Someone on the team counted words or pages (scans and phone photos included), looked up the rate for the language pair and the certification type, worked out the express surcharge and replied, often hours later. By then the client had frequently gone cold.
Once a job was sold, the operational work began. Jobs were handed out to more than 100 translators over chat. Quality control, deliveries, translator payouts and corporate receivables all lived in Google Sheets.
The problems to solve:
- Manual quotes that took hours, with scanned documents impossible to count by eye.
- More than 100 translators coordinated through chats and spreadsheets.
- Three currencies, seven payment methods and mandatory e-invoicing in Colombia.
- Pricing and availability rules that change by language, certification, client and urgency.
Why off-the-shelf translation software didn't fit
There are translation management systems and generic quoting widgets on the market. The issue was not a lack of software; it was that SWAP's business rules sit at the intersection of several things those tools treat separately.
A single quote depends on the language pair, the certification type, the urgency, the volume and, for corporate clients, a negotiated rate card. The same price has to be produced identically on the public website, inside the client portal, in the B2B portal, by a WhatsApp agent and when staff edit an order. Payments have to land in COP, USD or AUD through seven different methods, and every sale in Colombia needs an electronic invoice with DIAN.
Then there is production: real daily capacity per translator, freelancers who compete for offers, a QA pool working in shifts, notarization, apostille and courier dispatch. Stitching five or six SaaS products together would have meant five or six versions of the truth. SwapAI takes the opposite approach: one transactional core that every channel calls.
What we built: an AI quoting system that became twelve systems
We shipped the AI quoting MVP in six days and have kept building in phases on the same foundation, from the moment a client uploads a document to the moment the translator gets paid. This is ongoing work under our web and app development and AI and automation practices, with integrations and maintenance alongside.
Instant AI quoting from scans and photos
The public quoter accepts up to 50 files per quote. OCR reads scans and phone photos, and the system detects blank pages, duplicates and protected PDFs before pricing. The live price reflects the language pair, the certification, the urgency and the volume, so the client sees a number and a delivery date on screen instead of waiting for a reply.
A WhatsApp sales agent that never invents a price
Many clients still start on WhatsApp, so the agent lives there. It reads the documents clients send, quotes with the same engine as the website, detects payment receipts and hands over to a human only when it should. Rates, coupons and express availability are injected live into each call rather than written into the prompt, which is why the bot cannot make up a number.
Payments and e-invoicing in three currencies
SwapAI handles seven payment methods (Wompi, Stripe, PayPal, OpenPay, DeUna, Zelle and bank transfer) in COP, USD and AUD, with automatic DIAN e-invoicing through Siigo. Payments are financially idempotent: a payment creates exactly one order even if the webhook fires twice, and a backup cron covers the case where it never arrives.
Client and B2B portals
Clients can edit a quote before paying. Corporate accounts get negotiated rate cards, their own SLAs, prepaid or credit terms, statements per currency, aging and partial payments. Accounts receivable stopped being a spreadsheet and became part of the same database.
An automatic assignment engine
Every paid order is split into work units. In-house translators receive work according to their real daily capacity: quota, cut-off, schedule, absences and what they already delivered that day. Freelancers receive simultaneous offers, and the first to accept wins. The engine runs every two minutes, with locks against double assignment.
Quality control, translator portal and payouts
An auto-assigned QA pool works in shifts and reviews deliveries with a rubric; findings are attributed to the translator, and delivery alerts fire before anything reaches the client. Translators get their own portal with watermarked previews of offers, file delivery and deadline reminders. Monthly payment statements are generated atomically in SQL, numbered and versioned.
Fulfillment and management reporting
Notarization, apostille and courier dispatch are tracked in the same system. Management reports break down conversion by channel, translator punctuality and demand by language.
Under the hood
SwapAI looks like a clean web app, but by our own estimate about 70% of the effort went into backend, data, integrations and AI. The transactional core has 119 Postgres tables, all under row-level security, 265 SQL functions with business logic, 52 database triggers, 241 API endpoints, 7 signed inbound webhooks and 18 scheduled engines. Staff with combinable roles, clients, corporate accounts and translators share one database, and each sees only their own data through a 35-permission role matrix, signed short-lived file URLs and IDOR protection on every download.
AI is inside the workflow rather than bolted on: Cloud Vision OCR has run on 2,470 documents, Gemini answers on WhatsApp with a fallback model, retrieval runs over SWAP's knowledge base, Whisper handles voice notes, and an extraction cache avoids paying twice for the same file. Every call is logged for cost control.
We build this way with what we call Hybrid AI Coding: a senior WebMinds engineer directs AI coding agents (Claude Code) that implement, and an independent QA agent reviews every change before production. Architecture, judgment and client decisions stay human. That is how 592 commits and 212 migrations shipped in about six and a half months.
If your operation also lives in chats and spreadsheets, get a free assessment and we will map which parts can be automated first.
Timeline
- March 27, 2026: project kickoff.
- April 2, 2026: MVP in production with automatic quoting, an admin with roles and an 8-stage CRM pipeline.
- June 2026: Phase 2 adds the client portal, integrated payments and Siigo e-invoicing.
- August 12, 2026: the WhatsApp AI agent starts serving clients.
- September 2026: Phase 3 adds B2B pricing and portal, suppliers, production, QA and the translator portal.
- September 18, 2026: the order becomes the central record, replacing spreadsheets.
- October 6, 2026: the automatic assignment engine is switched on.
- Next: Phase 4 in planning, with multi-user B2B accounts, multi-language quoting, management reports and AI translation after payment.
Results
From hours to seconds
Any client uploads documents, scanned or photographed, and gets a price and delivery date instantly, on the web or on WhatsApp. More than 1,600 quotes and over a million paid words so far, with the same pricing engine on every channel.
An operation that assigns itself
Since October 6, 2026, every paid order reaches the in-house translator with free capacity or is offered to freelancers in seconds, goes through QA with shifts and a rubric, and is delivered by secure link. The spreadsheets and coordination chats are gone.
One system, end to end
Quoting, payments in 3 currencies and 7 methods, e-invoicing, production, QA, physical fulfillment, B2B receivables and payouts to 100+ translators all run on one auditable database with role-based permissions. You can see the screens and the full breakdown in the SWAP Translators case study.
FAQ
Can an AI quoting system price scanned documents and phone photos?
Yes. SwapAI runs OCR on scans and photos, detects blank pages, duplicates and protected PDFs, and then prices by language pair, certification, urgency and volume. It accepts up to 50 files per quote.
How do you stop a WhatsApp bot from quoting the wrong price?
The bot never calculates prices on its own. It calls the same pricing function as the website and the portals, and rates, coupons and express availability are injected live on every request instead of living in the prompt.
Does a translation agency need custom software, or is a translation management system enough?
If your pricing is simple and you work in one currency, a packaged tool may be enough. SWAP needed one engine shared by five channels, three currencies, seven payment methods, DIAN e-invoicing and capacity-based assignment for 100+ translators, which is where a custom core paid off.
How fast can a project like this reach production?
In SWAP's case, the AI quoting MVP was in production six days after kickoff. Everything else (payments, portals, the WhatsApp agent, QA and automatic assignment) was added in phases on the same foundation.
Do you work with companies in the US and Latin America?
Yes. WebMinds is based in Atlanta and works in English and Spanish with clients in the United States and Latin America. SWAP itself operates in Colombia, the US, Ecuador and Australia. For another Colombian example of a regulated workflow turned into software, read the IRS VIAL forensic platform case.
Ready to stop quoting by hand?
SwapAI started as a six-day quoting MVP and became the system SWAP runs on. For a deeper look at the platform, visit the full SwapAI case study, or see how we built a Colombian freight platform for ANT Mobility. Our integrations and platforms work covers the payment and e-invoicing side.
Get your free assessment or book a call with our team.