Prabandhak
India's first AI translation marketplace — preserving formatting global engines couldn't.
India's first AI-powered translation marketplace
Overview
We were a technology company, but the inbound kept asking for one thing: translate our documents. We were being compared to language service providers — so I built the product that turned that competition into customers.
Prabandhak became India's first AI-powered translation marketplace — built end-to-end over a year of beta and three revamps, shaped by real translators before it went public.
01 — The bet
Turn the competition into customers
Freelance translators, language service providers, and clients who needed documents translated — all on one marketplace. Instead of competing with the agencies, Prabandhak gave them the tooling and the demand, and took a place at the centre of the flow.
02 — The workspace
A translator's cockpit
A computer-aided translation tool built for Indian languages — a split-view workspace where every assist a translator could need sits one keystroke away, and nothing breaks the document underneath.
Inside the workspace
Eleven modules, one seamless flow
Swalekh typing
Native Indic typing — inscript + phonetic — built in, with personalised suggestions.
Indic spellcheck
SymSpell, adapted for Indian-language scripts.
Indic normalizer
Canonicalises inconsistent Unicode and encodings.
Font converter
Legacy font ⇄ Unicode conversion for ingestion.
Automatic QC
Live checks — length, punctuation, and styling shown as abstracted tags.
Translation memory
Reuse with search and named-entity substitution.
Term base
Enforced terminology consistency across a project.
NMT suggestions
In-line machine-translation, every edit logged and scored.
Review workflow
Translate → proofread → edit → export, with roles and permissions.
Collaborative workspace
Many translators and reviewers on one project.
Realtime analytics
Dashboards, milestones, and NMT-efficiency tracking.
03 — The engine
Suggestions that learned from every edit
Prabandhak suggested translations inline — then learned from what came next. Every translator edit was logged and scored against the suggestion by edit distance and BLEU. Those corrections fed an automatic data-preparation step that trained a seq-to-seq engine in-house — a flywheel plain-text engines never had, because they never saw the formatted, corrected truth.
04 — The differentiator
Format, preserved end to end
Global translation engines worked on plain text. Prabandhak kept the structure and styling of complex government documents intact — a 30+ format converter built from scratch, normalising rich files into one common format and rebuilding them perfectly across 11 Indian languages.
Trusted to sign off
At the Income Tax Department of India, senior IRS and IAS officials used Prabandhak directly to review and sign off on published content. I worked with them customer-first — turning their requests into fast iterations on ease-of-use, so the tool fit how they already worked.
- Income Tax Department of India
- GeM portal
- Infosys (system integrator)
- Translation agencies
Everyone else translated plain text. Prabandhak kept the formatting.
The bigger bet
Every job quietly collected something rare — real, formatted, parallel translation data, the seed for a formatting-aware Indic engine plain text could never match. In 2019, Anuvadak took the same engine and made it an API: a JavaScript snippet that localised a website in place, translations appearing live as they completed.
Role
Conceived and built the product end-to-end — the marketplace, the CAT tool and its modules, the format pipeline, and the data strategy behind a rich-text Indic translation engine — revamping it three times on real user feedback before going public.