reddy@fremont:~$ whoami

Reddy Balaji
Madha.

I build

Backend developer with 4+ years shipping production systems for EMS and healthcare — from offline AI running inside ambulances to clinical data pipelines, dispatch telemetry watchers, and Snowflake analytics platforms.

open to opportunities Fremont, CA Software Developer @ Atian.ai
// 01 · about

Systems that keep working
when the internet doesn't.

Reddy Balaji Madha
status: building · Fremont, CA

I'm a backend software developer focused on production APIs, edge AI, and healthcare data systems. At Atian.ai I architect end-to-end systems in Java/Spring Boot and Python/FastAPI — NEMSIS 3.5.1 SOAP pipelines, AI-assisted patient care report generation, offline AI deployed on ARM64 devices inside ambulances, Snowflake-backed operations dashboards, and real-time Elasticsearch infrastructure.

Before this I spent two and a half years at Tata Consultancy Services doing automation and infrastructure reliability work — the foundation for how I build now: self-healing services, offline-first design, and test suites that don't need live dependencies. Currently pursuing an M.S. in Computer Science at Jessup University.

4+
years experience
2
live production platforms
180+
offline tests, zero live deps
0
internet needed at the edge
// 02 · stack

What I work with.

The tools behind the systems — backend first, with a deep edge-AI and healthcare-standards streak.

Backend Engineering

Production APIs and microservices — REST, SOAP, async workflows, schema validation, JWT auth, and XML/JAXB pipelines that survive vendor quirks.

Java · Spring BootPython · FastAPIaiohttp / asyncioREST + SOAPJWTXML / JAXB

Edge AI & IoT

AI that runs where the cloud can't reach — quantized models on ARM64 routers, cross-compiled containers.

Whisper.cppMedGemma 4BYOLOv8 · ONNXDocker buildx / QEMU39 MB edge footprint

Healthcare & Clinical AI

The unglamorous standards that make medical data move.

NEMSIS 3.5.1HL7 v2FHIRICD-10 · RxNorm · SNOMED CTGPT PCR extraction

Browser Automation & Reverse Engineering

Production-scale automation against systems that were never meant to be automated — undocumented protocols, legacy UIs, safety-gated write paths.

Playwright / CDPSignalR reverse-engineeringnoVNC + Xvfbsafety-critical design

Languages

01Java
02Python
03SQL
04Bash
05JavaScript

Data

Snowflake · Snowpipe
Elasticsearch
OpenSearch
DynamoDB
SQLite · Kafka

Cloud & Infra

AWS · EKS · ECS
Kubernetes · Kustomize
Docker
GitHub Actions
Linux ops

Reliability

self-healing watchdogs
retries · backoff
health state machines
structured logging
offline test harnesses
// 03 · experience

Where I've worked.

Atian.ai (HealthyVibes.ai)

Dec 2024 — present · remote
Software Developer

Edge AI inference platform — ARM64 · Whisper.cpp · MedGemma · YOLOv8

  • Deployed edge AI on Ericsson R1900 (ARM64, 4 GB RAM) — Whisper.cpp at 14 s per 30 s of audio, YOLOv8 at 750 ms via ONNX Runtime
  • Built the automated encounter pipeline: YOLO triggers start/stop, Whisper transcribes, MedGemma 4B generates SOAP notes and PCR reports across a 17-task NLP pipeline
  • Cross-compiled Docker containers for ARM64 with buildx/QEMU; stress-tested 1,015 transcriptions at 103/hour with zero crashes

EMS intelligence analytics — Snowflake · FastAPI

  • Built a multi-tenant analytics platform for live ambulance districts — response times, dispatch lags, fleet GPS, billing revenue, AI documentation scoring
  • Designed the Snowpipe ingestion pipeline powering real-time KPI dashboards across operations, clinical intelligence, and revenue

NEMSIS 3.5.1 AI-powered submission pipeline — Java · Spring Boot · GPT

  • Built the end-to-end NEMSIS 3.5.1 XML pipeline: ASR → GPT PCR extraction → ICD-10/RxNorm/SNOMED mapping → XSD-compliant SOAP submission to state and vendor endpoints
  • Reduced vendor rejection rate via dynamic empty-section suppression and two-pass prompt engineering; deployed on Kubernetes (3-replica)

Clinical audit, QR handoff & infrastructure

  • Python audit pipeline with Sentence Transformers for confidence-scored consistency checks between EMS transcripts and clinical reports
  • QR-based JWT clinical handoff — zero-login ED patient access with 8-hour token expiry and patient-scoped security
  • Diagnosed Cassandra quorum loss and Kubernetes pod networking failures on AWS EKS; managed HL7 v2/FHIR ingestion services

Tata Consultancy Services

Feb 2020 — Aug 2022 · Bangalore, India
System Administrator — Automation & Infrastructure

Automation & tooling

  • Designed internal automation tools in Python, Bash, and PowerShell to eliminate recurring manual work and streamline operational workflows
  • Developed SQL-based diagnostic scripts to investigate production incidents and help application teams restore service faster

Infrastructure & reliability

  • Built monitoring and health-check scripts across Linux, Windows, and macOS for proactive detection of reliability issues
  • Worked production incidents with app teams — log analysis, network diagnostics, container-level troubleshooting
// 04 · work

Things I've built.

Live production systems, safety-critical automation, and one meditation on civilization. Click any tile for the full case study.

Offline Edge AI architecture
flagship · atian.ai · fully offline

Offline Edge AI Medical Intelligence

Whisper.cpp + MedGemma 4B running on edge hardware inside an ambulance — real-time clinical documentation with zero internet.

1,015
transcriptions — 10 h stress test
0
internet required
39MB
edge footprint
open case study →
EMS Intelligence Analytics Dashboard
atian.ai · live in production

EMS Intelligence Analytics

Multi-tenant Snowflake platform for live ambulance districts — six tabs of real-time operations, clinical, and revenue KPIs.

open case study →
reliability engineering

CAD Telemetry Watcher

Reverse-engineered a proprietary dispatch feed; self-healing watcher streaming ambulance status changes 24/7.

38→100%
dispatch capture coverage
open case study →
safety-critical automation

ePCR Auto-Fill Agent

Safety-gated browser agent pre-filling QA-validated patient charts — every clinical decision stays human.

287
fields · 12 chart pages
open case study →
A L O N E
"What would it take to make this — alone, with nothing, in your lifetime?"

A meditation on civilization, disguised as a web app. Live in production.

open →
// more builds
// 05 · blog

Notes from the rabbit hole.

Experiments, investigations, and what I learn while testing new technology.

Cropped YouTube Music page for Malia showing songs with hundreds of thousands of plays
July 2026 · AI, music & transparency

Can You Tell When a Song Is AI-Generated?

A random autoplay session led me to test how convincingly AI can imitate commercial music, and what listeners are told about it.

read the investigation
// 06 · contact

$ let's build something_

Open to backend, edge AI, and platform roles — or just a good conversation about systems that have to keep working offline.

Fremont, California 805-732-2532 linkedin/reddy-balaji-madha github/reddybalajimadha