Work

Things I built, and what changed because of it.

Each entry says what the problem was, what I did, and the measurable result. Where a project is still running, the numbers are from production.

01

AI-Powered Healthcare Platform

The problem. Healthcare teams need to work with large volumes of medical records, clinical documents, compliance workflows, and surgical coordination data, while maintaining strict security and access controls.

What I built. Engineered a multi-agent AI healthcare platform using FastAPI, PostgreSQL, Redis, AWS, Pinecone, and LLMs to automate medical-record retrieval, knowledge management, and surgical coordination. Implemented RBAC, audit logging, workflow orchestration, and AI-powered automation for secure clinical workflows.

Multi-agent AI HIPAA-compliant RBAC Audit logging LLM integration Vector search
FastAPI · PostgreSQL
Redis · AWS · Pinecone
LLMs · Python
02

Multi-Tenant SaaS Platform

The problem. Multiple construction enterprises needed to use the same platform while keeping their data completely isolated from other tenants.

What I built. Architected and shipped a multi-tenant SaaS platform with FastAPI, PostgreSQL, Redis, Celery, and Docker. Built modules covering HRMS, payroll automation, geofencing-based workforce tracking, Live chating with video calling system for internal users and course management, while standardising tenant provisioning for faster client onboarding.

Architecture. Designed schema-per-tenant PostgreSQL isolation so multiple tenants could run concurrently on the same database cluster without cross-tenant data leakage.

Multi-tenancy Schema isolation Tenant provisioning Background jobs Geofencing Project management HRMS Accounts and Audit Assets Maintanance
FastAPI · PostgreSQL
Redis · Jitsi · Celery · Docker
03

Real-Time Trading API Platform

The problem. Trading applications require reliable APIs for consuming market data and managing orders with minimal latency.

What I built. Developed RESTful APIs using Django REST Framework and integrated Zerodha Kite Connect for real-time stock-data ingestion and order management. Designed the backend around low-latency API communication and external trading-service integration.

Real-time data Make Historical Data Strategy Analysis Risk and Execution Management Order management REST APIs External API integration
Django · DRF
Python · Zerodha Kite API
02

Real-time chat across multiple workers

The problem. Messages were arriving only some of the time. The connection manager kept its client list in memory, so delivery only worked when sender and recipient happened to land on the same worker — roughly one time in four.

What I built. Moved every delivery path onto Redis pub/sub channels keyed by tenant and user, so any worker can reach any connected client. Along the way I found the disconnect logging was swallowing its own exceptions and fixed that too.

1-in-4 → 100% delivery Read receipts Reconnect handling
WebSockets · Redis pub/sub
FastAPI · Uvicorn
03

Self-hosted video meetings

The problem. Video calls dropped after a few seconds, and a hosted provider would have added a per-minute bill to every customer conversation.

What I built. A self-hosted Jitsi stack behind a custom domain with SSL, nginx WebSocket proxying, and long-lived connection timeouts. The dropped calls turned out to be a missing UDP rule on the media port — one firewall line, found by reading the media server's own logs rather than guessing.

Zero vendor minutes Custom domain + SSL Stable long calls
Jitsi · Docker Compose
nginx · AWS ALB & ACM
04

Document intelligence assistant

The problem. Answers about a private document set had to be trustworthy — a confident wrong answer is worse than no answer at all.

What I built. A retrieval pipeline that chunks by document structure, stores embeddings in Postgres, and returns answers with citations pointing back to the source page. A labelled evaluation set of question-and-answer pairs measures retrieval accuracy before and after every change to chunking or reranking.

In progress Cited answers Eval-driven
RAG · pgvector
FastAPI · Claude API
08

Multi-Tenant Microservices Platform

The problem. A growing SaaS system needs independently deployable services, reliable asynchronous communication, caching, and strict tenant isolation without creating tightly coupled components.

What I built. Designed and built a production-grade multi-tenant SaaS platform from scratch using microservices. Developed high-throughput REST services in Go, introduced Redis for distributed caching and cross-service rate limiting, and integrated Kafka for asynchronous communication between services.

Architecture. Implemented tenant provisioning, notification dispatch, and audit logging through event-driven workflows and managed database schema migrations using Atlas with CI/CD compatibility.

Microservices Go / Gin Apache Kafka Redis Event-driven Atlas migrations
Go · Gin · Kafka
Redis · PostgreSQL · Atlas
05

Production hardening and secrets

The problem. Database credentials lived in a file on the server, and services didn't come back cleanly after a restart.

What I built. Credentials moved into AWS Secrets Manager and are injected at startup through an IAM role attached to the instance — nothing sensitive on disk. The API, worker, and scheduler now run as grouped systemd services that start, stop, and restart together.

No credentials on disk Least-privilege IAM Restart-safe services
AWS Secrets Manager
IAM · systemd · Poetry
05

Automated PDF Reporting Pipeline

The problem. Report generation was dependent on manual processing and could take hours to complete.

What I built. Engineered an automated PDF extraction and dynamic reporting pipeline using PDFMiner, PyPDF2, and Jinja2. The system extracted information from documents and generated dynamic reports automatically, reducing report turnaround from hours to seconds.

PDF extraction Report automation Dynamic templates Document processing
Python · PDFMiner
PyPDF2 · Jinja2
06

EyeNav — Hands-Free Cursor Control

The problem. Traditional mouse-based interaction can create accessibility barriers for users who need hands-free computer control.

What I built. Developed EyeNav, an accessibility-focused cursor-control system using computer vision and facial/landmark tracking to provide hands-free interaction with a computer.

Computer vision Accessibility Hands-free control Landmark tracking
Python · OpenCV
MediaPipe
07

Large-Scale Web Data Pipeline

The problem. Collecting structured information from large numbers of web pages manually is slow and difficult to maintain.

What I built. Developed large-scale BeautifulSoup-based web-scraping pipelines to collect and transform web data for downstream analysis and data-processing workflows.

Web scraping Data extraction Automation Data pipelines
Python · BeautifulSoup
Data Processing

Your project next

What would you want on this page?

Tell me the outcome you're after and I'll tell you what it takes to get there.