Nov 2022 – Present · 3 yrs 4 mos

• Data Analysis and Anomaly Detection
• Prompt Engineering with Langchain LLM for quick log analysis & anomaly detection
• Generative AI, Langchain with OpenAI, VertexAI
• Log analysis and anomaly detection
• Developed and implemented a pipeline for processing and analyzing log data
• Identifies and reports network anomalies, quick RCA
• Content Generation
• Effectively leveraged Langchain with context
• Prompt Engineering for training LLM for domain specific custom training
• Generating realistic network traffic and synthetic data to strengthen network security, RCA and predictive maintenance
• Automating network configuration to help scaling and reduce human error
• Generating content such as user documentation, unit tests from source-code
• Image Generation using DALL-E
• Story Generation with GPT-3
• Narrative-driven Network Insights

Apr 2020 – Jun 2023 · 3 yrs 2 mos · Bangalore

• + Leading team of engineers, AI/ML models for embedded base-stations, cloud and 5G UPF core to solve complex networking, analytical and 5G cellular problems.
• NLP Log Analytics for proactive anomaly detection for 5G/V-RAN cloud systems
• Traditional AI
• Leading PoC effort to identify the entire workflow from data tagging, collection, cleaning, storage, training, analytics
• Investigate time-series data for metrics and logs using LSTM model
• Detection of complex inter-application issues within minutes
• Dynamic Power Management for 4G/5G Base-station: Leading team of 5, Research to Commercialization, patented, IEEE WCNC 2023
• Embedded base-stations are resource constraint (CPU, Frequency, Memory) devices with no ML acceleration, always running at a constant frequency even under off-peak hours with hardly any CPU load.
• Created framework to derive math polynomial representing ML inference model
• Optimized expression efficiently realizable on base-stations (and other embedded systems, IoT sensors) to predict CPU load and adjust CPU frequency accordingly to conserve power.
• Average Power Savings of 10-12%
• TCP Flow ML classification as elephant/mice and AQM bandwidth regulation in UPF core, research to commercialization, patented
• TCP handover prediction and bandwidth regulation in hi/lo BDP cells
• Edge Light Weight AI: Optimizing Simple Neural Networks for ML inference on embedded devices
• Reinforcement learning (RL) framework to efficiently and optimally implement simple NN & classical ML inference for embedded CPUs, taking <10% CPU cycles and <20% less memory compared traditional NN inf
• Makes ML inf ideas realizable on embedded edge and IoT devices
• Call Mute reduction with reinforcement learning based RoHC enabling in 5G RAN, patented
• NR PUCCH dimensioning using AI, patented

Dec 2015 – Apr 2020 · 4 yrs 4 mos

• + Machine Learning for routing/networking (on-device analytics): ML application to predict and root cause network failure scenarios.
• Conceptualization, architecture, design and implementation of models for Network Troubleshooting Analytics
• Created framework for data collection, representation and cleaning for network events
• Worked with various ML models for event classification, prediction and closed loop system
• Hackathon 2017 first prize in Machine Learning track
• https://www.tdcommons.org/dpubs_series/1816/
• + Graceful Insertion and Removal (GIR, Maintenance mode):
• Led the architecture, design and implementation
• Goal of this feature is to gracefully remove (and re-insert) a device (switch/router) from the network with minimal disruption to traffic
• Assurance app to provide global analytical view of the network undergoing GIR
• Hackathon 2018 first prize in DNA-C Assurance track
• https://www.tdcommons.org/dpubs_series/1224/
• + SDWAN
• TCP Proxy with BBR algorithm
• + Enterprise switching (Cat9k):
• Evaluation and prototyping with new forwarding ASIC
• Stackwise Virtual (SVL) serviceability
• + Machine Learning projects:
• Predicting Mean Time To Resolution (MTTR) and component for a bug (Team Coach)
• Strategies to determine whether performance requirements can be satisfied in a 4G/5G/WLAN Network
• https://www.tdcommons.org/dpubs_series/1385/

Aug 2005 – Dec 2015 · 10 yrs 4 mos

• Networking:
• TCP/IP stack for multi-core OCTEON, identified/resolved various bottlenecks in TCP scaling/performance, closely worked with HW team for HW offloads/acceleration, integration of various HW blocks to build fully RFC complaint, high performance, scalable multi-core TCP stack.
• SDN/Openflow offload (patented).
• Storage:
• Architecture work on NVMe based storage controller with flexible and extensible storage (several patents granted). Architecture of IEEE DCB features in OCTEON chip, RAID, iSCSI, CIFS acceleration.
• Wireless:
• In-depth analysis of cutting edge wireless technologies/3GPP specs (LTE, small-cell, WiMax), identified how to exploit parallel processing capability of multi-core CPUs for wireless control plane. MAC Scheduler hardware acceleration block (patented).

Jan 2001 – Jul 2005 · 4 yrs 6 mos

• TCP-Proxy, Wireless/wireline-profiled TCP stack, performance profiling and optimizations for Flexi product-line.
• MPLS LDP stack design and development.
• MPLS Redundancy/Fault-tolerance.