our best work. we deliver products that stand the test of time: from healthcare to live video to ecommerce, our work is everywhere.

We built real-time voice bots that talk with customers and handle calls on their own, understanding a range of Indian languages and accents so people can simply speak naturally.

We created a clear way to measure how well the bots understood callers, then used it to make them better, resolving 10% more calls end to end without handing off to a human agent.

We built an AI platform to review the quality of customer conversations across teams, working with 50k+ calls and around 10 million daily records, so everyone can see how their conversations are going almost as they happen.

We made this AI platform run faster and more reliably in production, and improved the visibility for teams to catch and fix problems before customers noticed.

We shaped technical decisions across teams, including how to build and ship AI reliably, and introduced good documentation practices.

We built provider and patient applications to view consolidated patient data and securely access and share health records.

We integrated with major US EHR systems such as Epic, eClinicalWorks, Athena, and Cerner, enabling smoother data exchange across healthcare providers.

We built an AI system that pulls context from clinical records and summarizes patient complaints, helping providers quickly understand why a patient is seeking care.

We developed standardized data processing systems that make health records easier to search, synchronize, and use in AI workflows across platforms.

We helped build privacy and security controls on AWS to support HIPAA compliance when handling sensitive patient data.

We owned the real-time voice stack end to end: bidirectional audio streaming, voice activity detection, barge-in, end-of-speech detection, multilingual transcription, intent classification, and live and post-call summarization.

We handled hard conversation problems in production, including interruptions, human-versus-robot caller detection, and mid-call goal changes driven by user input.

We also set up the evaluation framework and observability needed to measure call quality, debug failures, and keep improving the assistant after launch.

We built a data science platform that let teams build, test, and ship machine learning models faster, turning experiments into live features more reliably.

We built an AI system that reads driver documents with 90% accuracy, cutting manual paperwork and getting driver-partners onboarded faster.

We rebuilt the system that matches orders and generates price quotes across India. Uptime rose from the low-99% range to 99.9%, response times dropped from 1.5s to 300ms, and new data science models delivered 3% more fulfilled orders.

We redesigned the driver onboarding system by aligning product, design, analytics, and engineering teams around a shared model of the driver's journey.

We moved Porter's core logistics platform toward an event-driven design, making it easier for teams to build reliable features as operations kept changing.

We also advised Porter's leadership on engineering strategy, AI opportunities, and the architecture and practices needed to support their long-term product vision.

During our multi-year partnership, we led, and supervised product initiatives on the technical front, while collaborating with business, product, ops, design, research, and data science teams.

Our primary role was in engineering large scale, low-latency, stateful systems, and data pipelines for ML runtimes.

In addition, we established processes around designing software, project management, documentation and tech debt management.

Some of our major projects include rewriting and architecting the central driver allocation system in a few months, with a generative testing suite.

We're also happy to have introduced and trained multiple teams at Gojek in Clojure.

nilenso has worked with Simple since its inception in 2018, from building the first prototype, to launching the pilot solution, iteratively developing features, managing scale, as well as playing a pivotal role in Simple's expansion beyond India to other countries.

We designed the fully offline-first sync APIs that allow for the app to be used in rural areas with limited internet connectivity.

In addition, we built the dashboards that support both population-level administration by policy experts as well as more granular monitoring by health officials.

The dashboards were engineered for performance as well as accuracy using ETL pipelines as well as Russian doll caching.

The pilot was a real-time quiz that thousands of users would participate in. In addition to all the challenges that come with live video, we had to ensure that all participants received the questions at the same time.

Our solution involved a dashboard from where a producer could run the quiz, publish questions and answers to a massive live audience, and coordinate transitions from one question to the next.

We also built video transcoding into the platform using mimoLive and Wowza Cloud, so the live video stream from the quizmaster could be streamed to all participants.

The quiz and video were synchronized by stamping metadata onto the stream, self-identifying sync time, and compensating for server-side transcoding delays on mobile clients.

In order to load test the system, we built a Mob Simulator, that simulated hundreds of thousands of concurrent users.

The service was transitioned to an in-house team over months, while coaching them on Elixir, and OTP as well.

We initially consulted with Juspay on optimising the performance of their Postgres DB to handle high volumes.

Subsequently, we worked with them on building a PureScript framework to create mobile apps declaratively.

We also led the product and engineering initiative to build a next-gen dashboard to offer big-picture and action-oriented views for merchant PMs and customer support teams.

We engineered the data pipelines that aggregated data (including streaming data) from various sources.

While working on the dashboard project, we also created generic components that enabled other teams to quickly build dashboards for other products at Juspay.

Lastly, we partnered with Juspay to build and launch pilot ride booking and final mile delivery marketplace solutions based on an open protocol for location-aware commerce.

We built Staples' highly available experimentation platform with automatic failovers. The platform incorporated research from white papers from Google, Microsoft and Netflix.

We also built and maintained machine learning runtimes for Staples' email recommendation engine.

The experimentation platform included a simulation testing harness and framework that we wrote using Causatum, Simulant and Datomic.

Yet another project involved orchestrating a web crawler and scraper that would deal with unstructured datasets on the web and provide insights into categorisation as well as pricing on a daily basis.