I haven’t posted much about my job at Adalat AI. The team I’m part of builds dictation systems for the Indian court rooms, and initially, I thought it would be a straightforward extension of my PhD research. I was wrong.
Unlike academic work where success means beating SOTA on benchmark datasets, the reality at Adalat AI demanded something more: a deep understanding of India’s diverse courtroom dynamics, legal workflows, and the intricate linguistic landscape where regional languages and English constantly intertwine.
Real Courts, Real Challenges
What makes this work both exciting and challenging is our approach. Adalat AI ensures researchers and engineers truly understand the problem by immersing us in the environments we’re building for. We regularly visit courtrooms across different states, observe proceedings firsthand, and interact directly with judges and court staff.
This fieldwork revealed challenges no lab could have prepared me for. In Indian courtrooms, proceedings often flow through multiple languages—a judge in Odisha might switch between Odia and English, while witnesses in Kerala speak in colloquial Malayalam. Our AI needs to navigate these linguistic complexities while maintaining the precise meaning that’s so crucial in legal proceedings.
Building for Real Impact
As Kerala’s Judiciary moves toward implementing AI for transcribing court proceedings, we’re working tirelessly to create the most effective legal dictation system possible. Our team regularly conducts on-site training sessions to ensure smooth adoption.
During our recent visit to the Kollam district court, I observed a judge using our system during a witness deposition. Seeing how it handled the Malayalam-English transitions in real time revealed patterns our test datasets had missed. These insights immediately shaped our next development cycle. The AI resesearch team at Adalat AI invests in additional efforts to make the R&D systematic by defining tests that would really reflect the real world challenges.
Lawyers from different regions continuously test the system under various conditions, highlighting domain-specific challenges like regional legal terminology variations and procedural differences between court levels. Their practical perspective helps us build solutions that truly serve India’s diverse legal and linguistic landscape.
From Theory to Practice
Never in my wildest dreams did I imagine I would be training judiciary officials on AI tools. My research background prepared me for optimizing AI models. But my past classroom experience as a teacher proved to be invaluable when I had to explain technical concepts to veteran judges and court-staffs alike.
By reducing transcription time and improving accuracy in India’s many languages, we’re helping make justice more accessible and efficient for everyone across this diverse nation. And that’s an outcome more meaningful than any academic citation count.