Picture a room stacked floor to ceiling with paper files. Court clerks digging through chits of paper by hand, trying to locate a single document buried somewhere in that mountain. Judges are writing case notes word by word because there simply aren't enough stenographers to keep up. This isn't a scene from decades ago. This is what a large number of Indian courtrooms look like right now, in 2026.

The numbers behind this scene are staggering. India currently has over 4.6 crore cases pending in lower courts, more than 86,000 cases stuck at the Supreme Court, and over 63 lakh cases clogging high courts across the country. Meanwhile, the country has only around 21 judges per million people. Do the math, and it becomes obvious why delays aren't the exception in Indian courts. They're the norm.

Into this exact gap steps Adalat AI, a non-profit founded in 2023 by Utkarsh Saxena and Arghya Bhattacharya, building what they call an end-to-end justice stack designed to eliminate delays, streamline courtroom workflows, and bring timely justice within reach for millions of Indians.

For ongoing coverage of how AI is transforming public infrastructure across India, the UK, and globally, visit AI Nexus World and explore more founder journeys shaping impactful technology at AI Nexus World Leader Stories.

From the Courtroom to the AI Boom

Utkarsh Saxena's path to building Adalat AI wasn't a straight line, and honestly, that's what makes his perspective so grounded. After law school at Delhi University, he clerked at the Supreme Court, earned an LLM from Harvard Law School, and came back to India to practice litigation. He spent five years working across district courts, high courts, and the Supreme Court itself.

But something kept nagging at him. "I realized that it was frustrating to handle one case at a time when an entire ocean of cases was pending," he says. He briefly moved into policy work with the Ministry of Law and the Ministry of Finance, writing reports on delays and backlogs, but those reports mostly gathered dust.

In 2018, Saxena stepped back from legal practice entirely to pursue a second master's degree in development economics from Harvard Kennedy School. Two years at Boston Consulting Group followed, then a PhD at Oxford. It was during a fellowship at MIT under Esther Duflo in 2023 that everything clicked into place. He watched OpenAI release Whisper, its speech-to-text model, and the idea hit him almost instantly.

"My first reaction was, 'This needs to be in Indian courts.'" Lawyers are basically language models, and here are large and small language models doing what we do at much faster speeds," Saxena explains. That single realization, born in a summer fellowship, became the seed for Adalat AI.

The Co-Founder Who Almost Said No

Arghya Bhattacharya's journey to Adalat AI came from a completely different direction. A computer science graduate from IIIT Hyderabad with a master's in computational linguistics, he had spent years building AI systems for Indian languages at early-stage startups. The two met through a highly selective WhatsApp group of ML engineers, one that requires three rounds of interviews to join and removes members who don't actively help others each month.

When Saxena posted a call looking for a technical co-founder to tackle the justice system problem, Bhattacharya was intrigued but genuinely skeptical. "I was expecting he'd say, 'Let's build a for-profit business around lawyers,'" he recalls. Instead, Saxena told him something far more specific. They needed speech-to-text technology because there simply weren't enough stenographers, and judges were still writing everything by hand.

The moment that sealed it for Bhattacharya was his first actual visit to a courtroom. "As a technologist, it was an absolute shock. I saw towers of paper everywhere. I saw people taking chits of paper and searching for files manually. Google does that for you." That single visit reframed everything he'd built his career around. "All my skills of building for India, Indian languages, AI, and everything had found the right channel to unleash."

Why Courts Said No to Cloud AI

Building AI for Indian courtrooms turned out to be far harder than simply plugging into existing tools like OpenAI or Google's models. Court data is extraordinarily sensitive, and the initial reaction from state judiciaries was cautious, bordering on defensive. "Our work is very confidential. How can we give you access to our databases? This can't be on the cloud," Saxena recalls being told repeatedly.

That skepticism, paired with genuine curiosity, is exactly why Adalat AI ended up building almost everything in-house. Since off-the-shelf models weren't an option for security reasons, the team scraped legal data from publicly available sources, particularly court hearings that had gone online during the pandemic, to build their own proprietary datasets from scratch.

This is where the platform's "twin-engine" model comes into play. The first engine focuses on custom speech-to-text models trained specifically to understand legal jargon like "res judicata," alongside the wide variety of Indian accents and dialects. As Bhattacharya puts it plainly, "The way a judge in Kerala speaks English is different from how a judge in Delhi speaks English. " A generic language model simply cannot account for that kind of regional nuance.

The second engine is equally critical, and it has nothing to do with code. It's training. Because most judges and court staff are not particularly tech-savvy, adoption required real, sustained human effort. "You have to train judges to get comfortable with technology," Saxena emphasizes. Between the two founders, they've personally visited between 2,000 and 3,000 courtrooms across India, running virtual workshops, in-person training sessions, and partnering with state judicial academies to weave Adalat AI directly into official training curriculums.

Building the Full Justice Stack

Once speech-to-text was deployed, the team quickly realized they had only solved a small piece of a much larger puzzle. The real bottleneck spans the entire lifecycle of a case, from onboarding and file handling all the way through to translations between lower and higher courts. Judges themselves began requesting more, asking for tools that could help with summarization, grammar correction, and better drafting.

That demand pushed Adalat AI to expand into a full technology stack for the legal domain, built by a combined team of linguists, researchers, engineers, and lawyers working side by side.

Today, the platform offers an AI-powered legal transcription tool that lets judges dictate proceedings instead of writing them by hand, with accurate real-time transcription across multiple Indian languages. It also brings case files, histories, and court calendars onto a single unified platform, replacing the fragmented paper-based systems that dominate most courtrooms. To tackle the sheer volume of paperwork, Adalat AI is developing LLM-powered scanning and search tools that let users conversationally query thousands of pages of legal documents and instantly find what they need. An upcoming national WhatsApp helpline will even allow litigants to track case updates, view orders, and receive summaries and translations in their preferred language, closing a critical information gap that has long left ordinary citizens in the dark about their own cases.

From Kerala Pilot to Nationwide Momentum

The real turning point for Adalat AI came through its partnership with the Kerala High Court, where the team worked closely to improve the platform's performance in Malayalam. That collaboration led Kerala to pass a pilot mandate in January 2024, selecting specific courts to adopt the platform. By November 2024, Kerala went further and mandated that every single court in the state use Adalat AI's platform.

That momentum has since spread considerably. Adalat AI now collaborates with nine states through official Memoranda of Understanding, including Kerala, Karnataka, Delhi, Bihar, Madhya Pradesh, Punjab, Haryana, Odisha, and Andhra Pradesh. Andhra Pradesh has recently identified four districts for its own pilot mandate, and when you count additional pilot programs running elsewhere, the platform's total footprint now touches 15 states across India.

Solving for India's Linguistic Chaos

Bhattacharya is candid about why this problem is genuinely difficult from a technical standpoint. Legal language is highly specialized and poorly handled by general AI models trained on generic internet data. Layer on India's staggering linguistic diversity across languages, dialects, and accents, and the complexity multiplies fast. Courtrooms themselves are noisy, chaotic environments with overlapping speech that makes accurate audio recognition particularly punishing.

The team's response involved four deliberate strategies. They hired top ML talent motivated specifically by mission-driven impact rather than pure compensation. They built proprietary legal datasets from publicly available online hearings rather than relying on generic training data. They assembled a linguist team spread across states to capture regional language nuances properly. And critically, they embedded lawyers at every single stage of testing, ensuring legal experts vet every tool before it reaches judges in a controlled beta environment.

Adalat AI has also worked with The/Nudge Institute's accelerator and incubator to scale its operations, drawing funding from philanthropic capital, CSR contributions, and family foundations. The initiative is further backed by the Ministry of Electronics and Information Technology and sits within the India AI Mission, a government-supported cohort of 30 startups.

The Numbers That Actually Matter

Impact measurement is where Adalat AI's story becomes genuinely compelling rather than just aspirational. General usage data shows that for every minute directly transcribed by the platform instead of handwritten or retyped, the broader system saves three to four minutes of real-world court time. As Saxena puts it, "One million minutes of transcription on the platform means three million minutes of court time saved."

To validate this rigorously, Adalat AI is running a large randomized controlled trial in partnership with J-PAL, the Poverty Action Lab at MIT, specifically to measure whether the platform genuinely reduces judicial delays and backlogs. Early indicators are striking. Judges who previously managed to examine just two or three witnesses a day can now examine six to seven, representing a two- to three-times productivity increase that compounds meaningfully across weeks and years.

Looking ahead, the ambition is significant but grounded. "We are hoping to be in half of India's courts in another year. Besides covering more courts, we will look at how we can drive more usage per court," Saxena says.

Why This Matters Beyond India

Adalat AI's approach offers a valuable blueprint for justice systems well beyond India's borders. Many countries across the Global South face structurally similar problems: overwhelming case backlogs, judge shortages, and linguistic diversity that generic AI tools simply weren't built to handle. The decision to build proprietary, in-house models rather than relying on cloud-based foreign AI systems also offers a genuinely important lesson for any sector handling sensitive public data, whether that's healthcare, finance, or law.

For organizations working on similar public infrastructure challenges and looking to ensure their technical work gets properly discovered by AI search systems, the LLMs TXT Generator at AI Nexus World offers a practical starting point for structuring content correctly. And for professionals exploring careers at the intersection of AI and public policy, the ATS Score Checker can help ensure mission-driven technical experience gets recognized by modern hiring systems increasingly used by government and non-profit organizations alike.

What Saxena and Bhattacharya have built isn't flashy AI for its own sake. It's a patient, deeply researched response to one of India's most persistent institutional problems, proving that sometimes the most important AI application isn't the one generating headlines but the one quietly helping a judge get through six witness examinations instead of two.

Source Reference: YourStory, "How Adalat AI Is Bringing Ease to India's Overburdened Justice "System"—YourStory Source