Portfolio - Nikhil Utane

Operational Measurements (OM) and Key Performance Indicators (KPI) data is important for monitoring OneCell deployments in the field. Historically, this was done by a legacy tool that had significant limitations and was expensive to maintain.

Despite limited resources, I envisioned and developed a cloud-native analytics solution called Chronos, designed to provide comprehensive Operational Measurements (OM) and Key Performance Indicators (KPI) data for OneCell deployments. Chronos uses InfluxDB as a time-series database and is entirely hosted on AWS, offering interactive dashboards, reporting, real-time alerting, and advanced features like AI/ML-based troubleshooting and anomaly detection. Since its implementation, Chronos has effectively reduced the day-to-day monitoring cost and effort while delivering significant direct cost-savings of approximately $500K annually, in addition to boosting productivity, improving customer satisfaction, and unlocking new revenue opportunities. It is currently running in production, supporting both 4G and 5G deployments.

OneCell devices generate a huge amount of text-based logs. A single device can generate a few GBs worth of logs in a day. Earlier, there was no system available to store and query these logs efficiently.

Atlas is a centralized log analytics solution that is powered by Elasticsearch. With pre-built dashboards and queries, Atlas has significantly increased developer productivity by streamlining the process of searching logs. The solution utilizes a robust data pipeline that leverages Kafka and Logstash, running on AWS EKS. Additionally, it also includes a homegrown framework that automates complex stateful analysis. Together, these features enable efficient log data processing and analysis, reducing the burden of manual analysis and allowing for faster issue resolution.

The OneCell system is designed for high performance and low latency. However, deploying and managing the system can be complex due to its use of Kubernetes, which requires a certain level of expertise to manage effectively.

To simplify real-time monitoring of OneCell deployments, I created Themis, a single pane of glass observability solution. Themis is built using open-source components, such as, Thanos, Prometheus, and Grafana, offering multi-cluster monitoring and real-time alerting capabilities. With Themis, users can access a wide range of application, system, and Kubernetes metrics, and the solution also integrates with Chronos and Atlas to provide a one-stop-shop for monitoring and recovery needs.

Themis-Lab is a stand-alone version of Themis, which effectively manages and tracks the complete lab inventory while offering real-time utilization visibility.

It is a highly customizable solution that can be tailored to meet specific needs. It combines the static server data with real-time information to get accurate view of server utilization. This product has saved over $300K (and continues to do so) by identifying under-utilized or unused servers.

I led the DevOps effort to build a comprehensive & robust CI/CD pipeline for OneCell 5G & DAS product lines. I also led multiple other DevOps initiatives & built new tools to reduce inefficiencies and improve productivity, (for e.g. reducing the build time by 80%, using Tilt for developer environments & Rancher for Kubernetes cluster management).

In addition, I also brought in higher levels of automation using tools such as Ansible, Terraform, Robot framework to automate various manual & error-prone activities; These included setup bring up, install/upgrade, device configuration, release engineering practices, test automation etc

I have played a key role in driving & supporting the adoption of public cloud (AWS + Azure) along with other open-source, cloud-native technologies for the OneCell 5G program. This has allowed us to reduce hardware dependencies, deliver high-quality software, and increase agility & productivity while saving on both cost and effort.

I am also leading the Cloud-gNB project to run our 5G gNB solution in Azure cloud VM. This off-loads all functional testing to low-cost cloud VMs reducing the need for expensive physical servers.

I designed & executed a high-availability solution for the OneCell 4G controllers with 50% less time and cost. The original estimate for the project was 110 man-months and by cleverly designing using open-source components I achieved it in half the time with far superior quality.

The solution was appreciated for its innovative design and even received a patent recently. (Patent details: Title - CONTROLLER REDUNDANCY IN A CENTRALIZED RADIO ACCESS NETWORK. U.S. Patent No. 11,595,831 · Issued Feb 28, 2023) The project also required me to battle stiff opposition from some quarters due to the radical approach.

This one needs a special mention because of my unique ability to sift through gigabytes worth of packet captures in various forms (TCP, UDP, encapsulated, encrypted, fragmented, on-the-wire, over-the-air), day in and day out, to identify root-cause for the multitude of critical customer-impacting throughput-related issues.

A generative AI application to help prepare for behavioral interviews.

Developed using ChatGPT and Streamlit.

A generative AI application to answer questions about me.

Developed using LangChain, Next.js, Supabase and deployed on Vercel.

This was an attempt to highlight the growing polarization on Twitter between left-leaning and right-leaning handles. I scraped Twitter data from well-known LW & RW handles and trained a GPT-2 model to generate random tweets from a prompt. The results were remarkable.

The model was trained on Google Colab and deployed on AWS.

This is a sentiment analysis app that compares the two popular architectures of that time, BERT and ULMFiT.

The app is deployed on Google App Engine with Flask as back-end.

This is an image recognition app that identifies from the image whether the food is healthy or junk.

The model uses CNN, resnet50 to be precise and is deployed on Heroku.