KARSTEN MIRANDA
A team of students from Padre Conceicao College of Engineer ing (PCCE) has developed an artificial intelligence platform that allows us ers to interact with AI in Devanagari Konkani through voice and text — a project that has now won first place in the Software Category at Technix 2026, a national-level intercollegiate project competition held at Don Bo sco College of Engineering, Fatorda. The award-winning project, Konka ni Companion, was developed by Adrian Josephy Fernandes, Wendel Domnic Fernandes, Braise John Fer nandes and Saihaan Shaik Muzawar of the college’s Electronics and Com puter Engineering Department un der the guidance of faculty members Avita A. Lotlikar and Tina Vaz. The project grew out of what the students described as a simple ques tion: why should someone need to know English to benefit from artifi cial intelligence in 2026? Konkani, one of India’s 22 consti tutionally scheduled languages spo ken by more than two million people across the Konkan region, remains un derrepresented in modern AI systems, the team said. According to the stu dents, this often results in inaccurate responses, broken grammar and un intended switching to other languages, creating barriers for native speakers seeking to use emerging technologies. “Access to AI should not depend on proficiency in Eng lish,” the team told Herald. “Our target user was never a tech-savvy English speaker.
If you enjoyed reading this article, you can help support our journalism by adding OHeraldo as a preferred source.
It was a Goan homemaker, a village elder or a first-generation college student for whom language often becomes a barrier to information.” The students said they wanted to create a platform that treated Konkani as a primary language experience rather than an afterthought. “We wanted to build something that treats Konkani as a first-class language rather than simply adding a translation layer to an English-first system,” they said. One of the biggest challenges in developing the platform was obtaining sufficient language data for a language with limited digital resources. The breakthrough came when the team approached Goa University’s Vidyapaathi Lab, which works on language technology initiatives, and presented its concept. Through the lab’s Konkani Daan initiative — a crowd sourcing programme that collects spoken and written Konkani contributions from members of the public across Goa — the students gained access to community-generated language data that became the foundation of the project’s speech-recognition system. Built through the voluntary contributions of ordinary Konkani speakers, the dataset enabled the team to train the platform using authentic spoken language. “Without Konkani Daan, building a speech model capa ble of understanding spoken Konkani in real-world settings would have been extremely difficult,” the students said.
Konkani Companion functions as a multimodal AI sys tem capable of processing both voice and text inputs. Us ers can speak naturally in Konkani, after which the plat form converts speech into text, processes the query and delivers responses in Konkani. According to the team, the system operates through a translation bridge that converts Konkani input into Eng lish for processing before translating the response back into Konkani. This allows users to interact entirely in their mother tongue while the processing takes place behind the scenes. The platform also incorporates a custom-built Devan agari keyboard that allows users to correct transcription errors before queries are processed. The students said the feature was developed after recognising that speech inputs can sometimes be affected by pronunciation differ ences, dialectal variations or unfamiliar names. Unlike standard keyboards, the interface supports script-specific Devanagari characters, enabling users to make corrections without switching to another input method. Beyond voice interaction, the project includes a docu ment-intelligence feature designed to help users engage with Konkani-language documents. Using optical charac ter recognition (OCR) technology, the platform can extract text from uploaded articles, notices and PDF documents before analysing the content. According to the developers, a user who receives a gov ernment circular, community notice or newspaper article in Konkani can upload the document, ask questions about its contents and receive answers in both Konkani and English. In effect, the platform allows users to have a con versation with a document’s contents rather than simply reading it.
The students said the broader objective was to ensure that language does not become a barrier to accessing in formation and digital tools. Competing against entries from institutions across the country, Konkani Companion impressed judges with its combination of technical innovation and practical rel evance. The project eventually secured first place in the Software Category, bringing national recognition to both the team and the college. For the students, the award represents recognition of a project rooted in the belief that emerging technologies should be accessible in the languages people speak every day, including those that have traditionally received little attention in the global AI ecosystem.
