July 2024
navya bijoy
building scalable backend systems and robust infrastructure
I'm Navya, a Software Engineer building at the intersection of AI/ML and backend infrastructure, from predictive self-healing platforms to integration gateways and everything in between. Explore my work below :)
what i do
- Working with their AI team on Computer Vision Projects.
- Owned end-to-end feature delivery for a customer-facing production application serving 5,000+ daily active users, shipping Java-based backend services in Agile sprints.
- Optimized system reliability by engineering JUnit & Mockito test suites, expanding code coverage by 15% and preventing critical production bug leaks.
- Streamlined operational workflows by designing rule-driven automation in Java, reducing manual handoffs by 10% and increasing system throughput by 13%.
- Delivered a scalability-focused Proof of Concept (POC) simulating 10,000+ concurrent requests to validate high-availability architecture for a critical release milestone.
- Designed and shipped a REST API backend product with role-based access control (RBAC), validating security via 50+ Playwright end-to-end tests across 4 user roles.
- Migrated 10GB+ of binary asset storage to AWS S3, improving response latency by 30% and ensuring 0% unauthorized access.
open source
- Cut Stargate scale-up wait time by enabling early readiness detection, building a sampler that checks backend health every 1s and promotes readiness after 5 stable reads.
- Delivered a production Rust feature to NVIDIA's nvcf repo, validated by 349 tests, shipping a stabilization detector with timeout fallback.
- Made warmup tunable across deployments, adopted in 6+ NVIDIA services, by exposing new CLI/Helm config options.
- Created and maintain invisible-notes, an open-source Electron app for screen-share-invisible sticky notes, growing it to 35 stars and 20 contributors.
- Reviewed and merged code for 25 pull requests, guiding contributor submissions and maintaining code quality across the repo.
- Grew adoption to 120+ downloads, driving feature direction and prioritizing community-suggested improvements.
- Integrated 7 Databricks MCP tools, enabling AI agents to query SQL Warehouses, Vector Search indexes, and Unity Catalog functions.
- Engineered a read-only safety/guardrail layer for LLM tool-calling, blocking destructive SQL operations.
- Resolved a live multi-session race condition via two-layer PID validation across Linux and Windows.
- Contributed Machine Learning models spanning data preprocessing, training, and evaluation.
- Finished in the top 0.7% of contributors, merging 9 PRs into high-traffic ML repositories.
education
- CGPA: 9.83/10
- Academic Excellence Award: Scored 10/10 SGPA in III, V, VIII Semester. Secured 3rd for academic performance in SRM.
achievements
- LeetCode Knight (1909): Ranked in the top 4% globally across 800k+ participants. Placed 63 in weekly contest.
- Google DeepMind Hackathon: Top 25 Team at Bangalore Hackathon (2026), selected from top 250 participants out of 4000+ applicants.
- Research: Presented "Hybrid Learning and Enhancing Strip Steel Surface Defect Detection Using Adaptive FusionNet Framework" at an International Conference.
- Hack Summit 5.0: Won "Best Scalable Idea" and a cash prize among 200+ participants.
what excites me
I'm drawn to problems involving distributed systems, infrastructure automation, and backend scalability. Right now I'm deep into learning Go and Kubernetes, building robust platforms that can handle complex workflows and self-heal during failures.
let's connect
Want to discuss backend systems, collaborate on projects, or just chat about tech? Feel free to reach out!
projects
- Developed a Kafka-streamed metrics pipeline feeding a PyTorch LSTM anomaly predictor, forecasting incidents 3–5 minutes ahead of threshold breach.
- Cut MTTR from 5+ minutes to under 60 seconds through automated predictive remediation.
- Tracked model versions and automated weekly retraining/drift detection via MLflow, provisioning self-healing infra across 3 Availability Zones with Terraform-managed autoscaling (2 to 10 replicas).
- Built a pre-deployment chaos testing platform that injects realistic tool failures to detect harmful agent behavior; categorizes failures across a 4-quadrant matrix (safe/degraded vs. honest/silent).
- Shipped 38 preset scenarios across 6 tool domains (payments, database, email, filesystem, HTTP, search) for out-of-the-box coverage.
- Engineered a production-ready TypeScript monorepo CLI with a YAML scenario DSL, stateful mock environments, replayable test bundles, and Markdown/JSON/HTML reporting for reproducible CI testing.
- Built an end-to-end system that detects silent API breaking changes (enum removals, unit shifts, nullability creep) through LLM-powered semantic probes and deterministic verification.
- Implemented an AST-based call-site locator with a 5-step verification pipeline, reducing false positives from 41 candidates to 6 verified findings.
- Handled real-world code patterns including destructuring, aliasing, and optional chaining for robust cross-codebase analysis.
- Architected a pluggable Go SDK enabling third-party API integration (GitHub, Slack, Stripe) with zero core-engine changes.
- Automated API documentation generation via runtime schema introspection, eliminating manual OpenAPI 3.x maintenance.
- Built a background schema-drift monitor flagging breaking upstream changes within 30 seconds.
- Developed an automated remediation system handling 6 alert types via Prometheus/Alertmanager webhooks, reducing MTTR from 5+ minutes to under 60 seconds.
- Provisioned scalable, self-healing infrastructure across 3 Availability Zones using Terraform.
- Configured autoscaling (2–10 replicas) with CI/CD-enforced validation on every deployment.
- Built a self-serve two-stage data pipeline that clusters demand data and ranks candidate sites across 17+ variables.
- Architected an async REST API / microservice with a custom WebSocket manager for streaming live scoring progress.
- Designed a two-tiered SQLite coordinate cache with dynamic weight redistribution to cut redundant API calls.
- Designed an asynchronous backend microservice where REST API services enqueue jobs into Redis for independent worker processing.
- Architected job lifecycle management with persistent state tracking via PostgreSQL/JPA and retry logic guaranteeing at-least-once delivery.
- Benchmarked 100 jobs/sec system throughput with horizontal scalability.
- Built a multi-user agentic memory system that creates persistent knowledge graphs from browser-captured activity, supporting 500K+ embeddings.
- Implemented secure multi-tenant architecture using JWT + API-key authentication and a FastAPI backend.
- Built a semantic search and recommendation engine using 128D embeddings for intent-aware product discovery with sub-10ms matching latency.
- Architected a full-stack marketplace using MongoDB Atlas and Supabase, decoupling 100% of binary assets for high performance.
- Built an AI-powered pantry tracking system with real-time Supabase storage and Gemini-based recipe generation.
- Trained a MobileNet model achieving 96% image recognition accuracy and deployed it to Hugging Face.