I’m a software engineer at Baker Hughes, where I build AI tools that take repetitive work off people’s plates while keeping them in control of the results. I’ve also worked on network analytics at Juniper Networks and published research on machine learning for security. I like taking an idea from a paper to something that runs reliably in production.
CODE RED'25, 4th of 1,000+ teams · ELCIA Next-Gen Top 10
Education
CGPA 8.87
B.E. Computer Science (Cybersecurity) · RV College of Engineering
Core Tools I Work With
Core programming
C/C++
Python
Java
TypeScript
JavaScript
SQL (query optimization, data modeling)
Frontend & web
React
Next.js
HTML5
Tailwind CSS
Backend & distributed systems
REST APIs
Microservices
Node.js
Flask
Redis queues
WebSockets
Concurrent systems design
Databases
PostgreSQL
MongoDB
Redis
pgvector
Firebase
SQL modeling
AI / ML & frameworks
PyTorch
TensorFlow
Scikit-learn
Transformers (BERT)
CNN / LSTM
Data pipelines
Telemetry & observability
Cloud & DevOps
Docker
Kubernetes
Microsoft Azure
AWS
GitHub Actions (CI/CD)
Linux / Bash
Git
Core CS
Operating Systems
Computer Networks
Data Structures & Algorithms
System Design
Multi-threading
Experience Highlights
Software Development Engineer
Baker Hughes
Jan 2026 - Present
Promoted from Digital Technology Intern to Software Development Engineer (Jul 2026) after shipping the hybrid BERT-CNN classification platform to production.
Designed and trained a hybrid BERT-CNN NLP model for automated document classification at 85% accuracy, adding human-in-the-loop validation that cut manual audit effort by 50+ hours a week.
Engineered a full-stack classification platform (Python, Flask, React, PostgreSQL) with automated message queues, scaling partner intake throughput 3x across regional enterprise teams.
Architected production microservices on Microsoft Azure App Service with Microsoft Entra ID RBAC and GitHub Actions CI/CD, cutting deployment cycle times by 40% under a zero-trust model.
Software Engineering Intern, Data & Analytics
Juniper Networks
Jul 2024 - Feb 2025
Engineered a high-throughput Python processing pipeline handling 1M+ daily network packets, enabling real-time monitoring and automated labeled datasets for security analytics research.
Built and statistically tuned a microservice classification platform reaching 98% accuracy on network service identification, lowering system latency by 25%.
Established automated unit testing and validation frameworks in an Agile R&D workflow, reducing dataset error rates by 30% while holding production SLA compliance.
Education
2022 - 2026
CGPA 8.87
RV College of Engineering, Bengaluru
B.E. Computer Science and Engineering (Cybersecurity)
Data Structures & Algorithms
Operating Systems
Computer Networks
System Design
Database Systems
Machine Learning
Applied Statistics
Leadership
Aug 2023 - Present
Event Management Lead, Google Developer Student Clubs (RVCE)
Led Tech Tank for 500+ students, owning event operations, technical infrastructure, and cross-team coordination for the GDSC-RVCE community.
Projects
Distributed systems build
Shipped
CoLab
Real-time Collaborative Editor
High-performance real-time collaborative text editor with conflict-free synchronization, live presence cursors, and deep versioning built around CRDT principles.
TypeScript
React
Node.js
Redis
PostgreSQL
CRDTs
+2 more
Show detailsHide details for CoLab
Problem
Enable instant, conflict-free collaboration with secure session handling, auto-saving persistence, and version history snapshots.
System work
Designed the Y.js and WebSocket architecture, engineered compressed snapshot persistence in PostgreSQL, and implemented JWT-based auth with session rotation.
Outcome
Shipped a secure, low-latency collaboration stack with a real-time observability dashboard for system metrics and active user sessions.
Full-stack cyber defense platform with real-time threat visualization and containerized IoT simulation for attack-response workflows.
C/C++
Python
PyTorch
Docker
Kubernetes
Redis
+1 more
Show detailsHide details for DefenSys
Problem
Give teams a safe environment to observe threats and validate automated defenses without touching production systems.
System work
Built the React dashboard and Flask services, containerized the simulation environment, and implemented Redis-backed queues for asynchronous defense tasks.
Outcome
Shipped a working prototype for real-time threat demos and published the research behind the approach.
Retrieval-augmented Q&A over mixed-format technical documentation, with hybrid search, cross-encoder reranking, and per-claim citation grounding so unsupported claims are visible rather than hidden.
Python
FastAPI
PostgreSQL
Qdrant
Redis
Celery
+5 more
Show detailsHide details for AskMyDocs
Problem
Answer questions from trusted documentation without letting the model invent plausible-sounding claims that no retrieved evidence supports.
System work
Built the hybrid retrieval path (BGE-M3 dense vectors in Qdrant fused with Postgres tsvector keyword search via reciprocal-rank fusion), reranking with a multilingual cross-encoder, and a claim-level LLM judge that labels each answer grounded, partially grounded, ungrounded, or refused.
Outcome
Shipped a FastAPI service with streaming query progress, background Celery ingestion, a RAGAs evaluation harness against a 60-question golden set with regression thresholds, and deterministic safety screening for prompt injection and PII.
This site. A single scrolling portfolio built with Next.js 16 and React 19, MDX-backed long-form content, and a typed content layer that keeps copy and layout cleanly separated.
TypeScript
Next.js
React
Tailwind CSS
MDX
Vercel
Show detailsHide details for PrabuWeb
Problem
Ship a fast, accessible portfolio that is easy to update without touching layout code, and that renders as static output.
System work
Designed the single-page scroll architecture with anchor-based navigation and a scroll-spy header, authored the design token system and component primitives, and added SEO via the Metadata API with JSON-LD, sitemap, and OpenGraph cards.
Outcome
Deployed as a fully static build with all content sourced from typed data plus MDX, so copy changes never require component edits.
