8+ years architecting full-stack systems. I specialize in production generative AI (RAG pipelines, agentic workflows, AI copilots), ultra-low-latency livestreaming, real-time interactive systems, and AWS cost optimization.
Currently building moderator and audience tools for large-scale live events at Nova Dynamic Media.
- 🤖 Production RAG & agentic workflows on AWS Bedrock
- ⚡ Real-time interaction at scale with AppSync + GraphQL
- 🎥 Sub-second latency streaming with Amazon IVS
- 💰 Cut AWS spend ~40% through pragmatic architecture
- 🧰 Also comfortable across Python, PHP, Vue.js, and MongoDB/PostgreSQL
- 📍 Based in Kathmandu, Nepal
serverless-orchestration-blueprint — a production-grade, deployable reference for long-running, parallel, pollable job orchestration on AWS (Step Functions, Lambda, S3, DynamoDB). Extracted from the real event-streaming platform I run in production. Typed, tested, and solves the problems that actually bite: the 256KB state limit (S3 workspace-pointer pattern), API Gateway's 29s timeout, and bounded parallel fan-out with backpressure.
Most of my work lives in private client codebases — this repo is the pattern I extracted to show how I actually build.
|
RAG pipelines on AWS Bedrock over vector knowledge bases — powering moderator copilots, live & post-event chatbots, livestream summarization, and an agentic workflow (generator → validator → reasoning model) for plain-language metric queries. |
Live Q&A, polls, surveys & quizzes on AWS AppSync + GraphQL + DynamoDB, serving large concurrent audiences across 50+ productions. |
|
Amazon IVS real-time streaming over WebRTC and adaptive HLS delivery with token-based encryption — plus a cross-region Bedrock failover that cut AI throttling failures ~68% under peak load. |
Serverless architecture as code (AWS CDK), CI/CD with AWS CodeBuild, and ~40% cost reduction via right-sizing and capacity tuning. |
-D97757?style=flat-square){kind=icon}
| Project | Description |
|---|---|
| serverless-orchestration-blueprint | Open-source, deployable AWS orchestration pattern extracted from production |
| AI Moderator Copilot | RAG system on AWS Bedrock generating, grouping & answering audience questions live |
| Mission Control | Unified moderator dashboard for live interaction, moderation & reporting |
| Real-Time Backend | Migrated live Q&A/polls/surveys from IoT Core/MQTT to AppSync + GraphQL for scale |
| Telemetry Pipeline | Kinesis Firehose to S3 + Redis/Valkey for live stats, processed via Glue + Athena |
Open to collaborations and senior Full-Stack / Cloud / AI roles