// hello, world — I'm

Sai Prakash

Currently Senior Software Engineer at Cubet · Kochi, India

See the highlights Get in touch Download resume (PDF)

01 About

Senior backend engineer with 7 years architecting scalable distributed systems, cloud-native applications and ML/AI infrastructure — building backends that bridge hardware, software and cloud.

Lately: production computer vision, RAG pipelines and agentic AI, taken from proof-of-concept to horizontally scalable production.

0years building systems

0API response time via query tuning & caching

0latency per request, via re-architecture

0360° video over cellular

Skip ↓

02 Highlights

Key skills, wired to the results they delivered.

        Skills × impact
60s → 25sone pipeline, re-architected
Skills, wired to impactEvery highlight below lights up the skills that made it happen. Seven years across edge, cloud and AI.

        Highlight 1 / 5
−58%latency per request
Re-architected a 3D vision pipelineFrom 6 containers + Celery/Redis to bare metal on a custom AMI: ~60s → <25s per request, stable at 3 concurrent GPU instances.
Python
COLMAP
OpenMVS
AWS ASG
Custom AMI
S3
GPU serving

        Highlight 2 / 5
<200ms360° video over cellular
Real-time VR streaming in C++WebRTC signalling with STUN/TURN, session and call routing, cross-compiled for Qualcomm Yocto Linux.
C++
WebRTC
STUN/TURN
Yocto
CMake

        Highlight 3 / 5
1 serviceper detection model
Industrial anomaly detectionEvent-driven microservices on Kafka ingesting live PLC telemetry, one serving service per model, live alerts over WebSockets.
Kafka
Microservices
PostgreSQL
WebSockets
ML serving
Python

        Highlight 4 / 5
1 GPUfully local
Agentic RAG, on-deviceLangGraph agents routing queries to chat, document or vector search, with OCR/VLM captioning, all running on one RTX 4070.
LangGraph
Vector DBs
OCR / VLM
Local LLMs
FastAPI
Python

        Highlight 5 / 5
−40%API response time
IoT backends, and a team of 4AWS IoT Core and MQTT device backends, Kinesis streaming and Kubernetes; cut API response times by 40% and led a team of four.
AWS IoT Core
MQTT
Kinesis
ESP32
Kubernetes
Docker
AWS Lambda

      

Skip ↓

03 Featured builds

Six systems, one constellation. Scroll to fly through them.

Build 1 / 6 · Myelin Foundry · Computer vision infra

2D → 3D Point Cloud & Object Fitting

before

60s

after

60s

6 containers + Celery/Redis → bare metal · −58% per request

Photos in, dense point cloud out: reconstruction, segmentation and volumetric fitting (will the luggage fit the boot?), autoscaled on GPUs.

COLMAP
OpenMVS
AWS ALB/ASG
Custom AMI
S3

Build 2 / 6 · Myelin Foundry · GenAI

Automated RAG with Agentic Routing

chatdocumentvector search

illustrative demo

Multi-source ingestion with page-level provenance and OCR/VLM captions; LangGraph agents route every query, all on one RTX 4070.

LangGraph
Vector DB
VLM
OCR
Local LLM

Build 3 / 6 · Myelin Foundry · Real-time systems

VR 360° Real-Time Streaming

999ms streaming latency

illustrative demo

The whole backend in C++: signalling with STUN/TURN, sessions and call routing, 1080p 360° video over cellular on Yocto Linux.

C++
WebRTC
Yocto
CMake
YouTube Live

Build 4 / 6 · Myelin Foundry · Industrial IoT

Industrial Anomaly Detection

0PLC events streamed

⚠ anomaly detected

illustrative demo

Fault-tolerant, event-driven microservices on Kafka ingesting real-time PLC telemetry from manufacturing equipment, with a dedicated serving microservice per detection model, PostgreSQL-backed REST APIs and WebSockets pushing live alerts.

Kafka
Microservices
PostgreSQL
WebSockets
ML serving

Build 5 / 6 · Myelin Foundry · AI for the shop floor

RAG Operator Assistant

illustrative demo

Operators ask out loud and get answers from Chinese and English manuals: speech-to-text, OCR, translation, semantic search, local LLMs.

Voice
OCR
Translation
Vector search
GPU

Build 6 / 6 · Personal project

Algorithmic Trading Bot · Nifty 50

illustrative chart

Stock and options analysis that suggests Nifty 50 trades, self-hosted with Portainer and shared over a Tailscale mesh.

Python
Options data
Portainer
Tailscale

Skip ↓

04 Toolkit

The full stack, layer by layer, from the edge up to AI.

Layer 1 / 6

Edge, IoT & Real-time

Where software meets hardware: ESP32s, Yocto builds, WebRTC links and PLC telemetry.

Edge computing
Yocto Linux
ESP32
WebRTC
STUN/TURN
PLC integration
Home Assistant

Layer 2 / 6

Data & Tools

Stores and tooling that keep systems fast and observable.

PostgreSQL
MySQL
Redis
Vector DBs
Time-series
Git
CMake
Sentry
Tailscale

Layer 3 / 6

Languages

Python across the stack, C++ for real-time systems.

Python
Django
Flask
FastAPI
C++
SQL
React.js

Layer 4 / 6

Backend & Architecture

Event-driven services designed to scale horizontally.

Distributed systems
Microservices
Event-driven
REST
Kafka
Celery
Low-latency

Layer 5 / 6

Cloud & DevOps

AWS and Azure, containerised and automated end to end.

AWS Lambda
EKS
EC2 / ALB / ASG
IoT Core
CDK
Azure Functions
Kubernetes
Docker
CI/CD · IaC

Layer 6 / 6

ML & AI Infra

Taking models from prototype to GPU-backed production.

COLMAP
OpenMVS
Segmentation
OCR / VLM
RAG
LangGraph
Local LLMs
GPU serving

Skip ↓

05 The Journey

The professional years: Bengaluru to Kochi. Scroll to fly the route.

tower height = time spentarc = the move

05 · role by role

Sep 2026 — Present now

Senior Software Engineer

Cubet ↗ · Kochi

Bringing distributed-systems, cloud and AI-infrastructure experience to Cubet's product engineering.

Aug 2024 — Sep 2026 ↺ boomerang

Senior Product Engineer

Myelin Foundry ↗ · Bengaluru

3D vision pipeline re-architected to bare metal: ~60s → <25s, with GPU-aware autoscaling.
Agentic RAG on LangGraph, fully local on one GPU.
C++ WebRTC VR streaming (<200ms) and Kafka anomaly detection.

Sep 2023 — Jul 2024

Software Engineer

Corestrat Labs ↗ · Bengaluru

Serverless trading backend on Azure Functions, with backtesting and risk management.
Write-heavy MySQL schema, WebSocket APIs, IaC and CI/CD.

May 2020 — Sep 2023

Product → Senior Application Engineer

Myelin Foundry ↗ · Bengaluru

Ran company IT alongside product engineering for the first year.
IoT backends on AWS IoT Core, MQTT and Kinesis; −40% API response time.
Led backend and AWS architecture and a team of 4.

Dec 2019 — May 2020 full-time starts

System Administrator · Contract IT Engineer

Opt IT Technologies ↗ · Bengaluru

Deployed on-site at Myelin Foundry from 19 Dec 2019; hired directly by Myelin in May 2020.
AWS and Azure infrastructure, hybrid Windows/Linux administration.

Next ★ the goal

Forward Deployed Engineer

On the ground with the people who use the system, owning it from first conversation to production.

Started forward-deployed: my first full-time role was on-site at a client, who then hired me.
Real-world AI: industrial telemetry, voice-driven troubleshooting, local LLMs.
Edge → cloud → GPU: end-to-end ownership, prototype to production, fast.

Skip ↓

06 Learning

Before the career: the long way round, 2010 to 2019.

pin = where I studiedarc = the movedashed = remote work

Skip ↓

07 Beyond the terminal

What I build when nobody's paying me to.

Off the clock · 1 / 4

Drone builder & aerial photographer

Hands-on hardware assembly and flight systems, plus freelance photography and drone shoots.

Off the clock · 2 / 4

Home automation

Designed and wired a full smart home: bulbs, sensors and cameras unified in Home Assistant.

Off the clock · 3 / 4

Pet-monitoring system

Camera-based monitor built with OpenClaw that tracked a cat's post-surgery recovery, with live Telegram updates.

Off the clock · 4 / 4

Freelance web developer

Designed and built custom websites end to end for independent clients.

08 Let's build something

Open to interesting problems in distributed systems, cloud and AI infrastructure.

[email protected] LinkedIn ↗

Kakkanad, Kochi · India