Jovin Nicholas | Software Engineer

About

Co-founded SevaTech, led a 10-person team to build and ship an Android app for an NGO, boosting event registration by 200%
Mentored 20+ students through Northeastern's Bridge to Calculus, teaching fundamentals of programming
Built agentic AI systems - LLM orchestration, RAG pipelines, and runtime tool-use, not just API wrappers
2.5 years shipping production code across Java/Spring Boot enterprise backends and Next.js/FastAPI product stacks
Loves debugging and fixing problems - SSO outages, WebSocket failures, undocumented APIs

What I Bring

How I add value.

Full-Stack Development +

Built a custom WebSocket adapter bridging a Next.js frontend to a Python FastAPI backend - because standard streaming APIs can't talk to custom Python backends.
Won Best Innovation at Hack NUACM 2025 - built and deployed a full AI platform in 24 hours.

AI + Product Engineering +

Chose RAG over fine-tuning at Health Innovators - grounding in live clinical data, not memorization. Achieved 90% accuracy improvement over GPT-4o baseline
Built an agentic NLP pipeline using Gemini function calling - cut clinical review from 15 min to 2-3 min per record
Holds Anthropic MCP and Oracle GenAI Professional certifications (2025)

Backend Systems +

Rearchitected a 7-step SOAP API checkout chain at LTM that had no retry logic and silently dropped failed pricing calls - drove a 25% transaction completion improvement for 5,000+ users
Volunteered to own a critical WSO2 SSO outage - traced Nginx proxy misconfiguration breaking OAuth2 callbacks, restored auth, unblocked go-live

Cloud + Infrastructure +

Deployed on AWS ECS Fargate across 4 AZs - ALB/NLB layer separation, zero-downtime CodeBuild CI/CD
Automated Docker image builds and ECR deployments via GitHub Actions pipeline, eliminated manual deployment entirely

Selected Work

Things I've built.

01 / 04

nuk.ai

Won Best Innovation at Hack NUACM 2025. Built and shipped in 24 hours.

Students spent time manually searching for a seat in the overcrowded library. I built a real-time occupancy platform that predicts crowding and lets students query availability in plain English.

Meta Prophet forecasts capacity from simulated IoT sensor streams - students see predicted occupancy, not just current
Gemini-2.0-Flash chatbot answers natural-language seat queries against live Supabase data
Deployed full-stack on Render within the 24-hour window, serving 3,000+ students

Next.jsFastAPI GeminiSupabase Meta Prophet

GitHub ↗

02 / 04

CatchMe

Production-grade AWS infrastructure for a real-time multiplayer game.

Real-time multiplayer requires consistent low-latency state across every connected client. I built the full backend and infrastructure to handle live player sync, high availability, and zero-downtime deployments.

Redis pub/sub syncs player state across containers in real time
Deployed on AWS ECS Fargate across 4 AZs - ALB for HTTP/WebSocket, NLB for Redis to minimize protocol overhead
CodeBuild CI/CD delivers zero-downtime rolling deploys, 50% faster than manual process

FlaskReact RedisAWS ECS Fargate DockerCodeBuild

GitHub ↗

03 / 04

JSON Schema Registry

Event-driven backend built to stay fast under write-heavy load.

Schema registries need fast reads for search and reliable writes for storage - but coupling them means write spikes hurt search performance. I decoupled them with async queuing so each concern scales independently.

RabbitMQ queues writes asynchronously so Elasticsearch indexing never blocks the write path
Full CRUD on Redis for low-latency schema retrieval, parent-child Elasticsearch indexing for search
OAuth2 security throughout; Spring Boot REST API design

Spring BootElasticsearch RedisRabbitMQ OAuth2

GitHub ↗

04 / 04

SafeDine

Turned messy city open data into a usable, accurate restaurant safety tool.

Boston's public health violation data is fragmented and full of duplicates - restaurants appear under multiple names and addresses. I built a pipeline to clean, deduplicate, and enrich it into something people could actually use.

MongoDB aggregation pipeline deduplicates records and enriches via Google Places API cross-validation
Achieved 98% data accuracy through entity matching and multi-source reconciliation
Full MERN stack with Redux + Thunk, improving client-side performance by 25%

Node.jsReact MongoDBGoogle Places API Redux

GitHub ↗

How I add value.

Things I've built.

nuk.ai

CatchMe

JSON Schema Registry

SafeDine

Let's buildsomethingtogether.

Let's build
something
together.