I’m Atharva Bagul. I build systems that keep working when the connection doesn’t.
WorkProductsContactView my CVLinkedIn ↗
Follow the packet: it leaves a device at dawn and reaches the data centre by night.
scroll to send it ↓
Before I built systems, I learned how they break.
I studied Computer Science and Engineering at Sandip University in Nashik from 2020 to 2024. Some of my most useful lessons happened outside class.
Remote internships with the Virtually Testing Foundation gave me hands-on practice with real failure modes in lab environments: auditing VMware environments, fixing IAM and S3 policy misconfigurations and AccessDenied errors, investigating incidents with CloudTrail and CloudWatch, and practising incident response with IDS/IPS, SIEM and malware-analysis tools.
I also mentored with Google Cloud Community India (2022–23) and the Google for Developers India Educators Program (2023), helping students learn Google Cloud, machine learning, generative AI and app development.
SunHacks 2022: first runner-up in the FinTech category, as a team.
Small cores, big questions.
I got curious about what runs underneath everything: the processor. I was first author on two papers about open-source RISC-V cores for low-power IoT devices: a study in JETIR and a survey on TechRxiv, both in 2024.
Since September 2023 I’ve been an individual member of RISC-V International’s Certification Steering Committee, joining its weekly meetings on certification standards.
Quantum computing pulled me in too: a research internship with QWorld studying stakeholder values in quantum technology, diplomas from the University of Latvia and Womanium’s Global Quantum Computing Program, and IBM Quantum awards from the 2023 Qiskit Global Summer School and Spring Challenge.
From pen and paper to something that flies.
In 2024 I joined the founding team of ANTIMATRIX (formerly SkyCircuit Innovations) with five batchmates, starting from pen and paper. For two years I was Lead R&D Engineer on a deep-tech UAV startup for defence and agri-tech, and wore every other hat too: soldering hardware, writing embedded Linux code, preparing proposals and pitches for VCs and government programs (iDEX, Startup India, MeitY), and building an outreach wing that formalised an MoU with the Sandip Foundation.
I worked on Grape, an integrated flight computer running embedded Linux for onboard edge AI and real-time telemetry, and on an autonomous long-range quadcopter and a vision-based aerial monitoring system with real-time object detection at the edge, from hardware–software integration through field testing.
Some nights meant debugging telemetry at 2 a.m. before a live field demo. In the air, a dropped sensor packet isn’t a log line; it’s a crash. That instinct for failure has stayed with me in everything I’ve built since.
Saved, sent, accepted, processed: four different promises.
The same question followed me from the air to the cloud: what happens when the connection fails? I built a set of projects to answer it properly, and to be honest about what each one proves.
EdgeOps keeps device readings in a local outbox, sends them through authenticated AWS ingestion into a queue and a worker, and protects current state from replays. I rehearsed a release, a rollback and a restore with Terraform and Jenkins. RestoreProof checks that an application still works after a restore, not just that the data came back.
Device telemetry that survives dropped connections: a durable local outbox, queued AWS ingestion, and a rehearsed release-and-rollback workflow.
ResultRelease, rollback and restore rehearsed; delivered artifact hashes matched every checkpoint.
AWS SQS · Lambda · DynamoDB · Terraform · Jenkins
Public source (MIT) · Controlled AWS deployment
Checks that an application still works after a restore, not just that the data came back.
ResultHealthy restore and the expected failure cases recorded; real S3 retrieval documented.
Python · AWS S3 · Docker · GitHub Actions
Implemented · Controlled workload · Public source (MIT)
Links a technical claim to the artifacts, checks and limits that back it up.
ResultPublic worked example: a broken and a corrected rollout, reproducible locally.
Evidence packs · Reproducible checks
Public pilot · Simulation authored with Codex
Collaboration that rejects stale saves without losing your unsaved draft.
ResultStale saves rejected with the draft kept; access and revision tests pass.
Revision checks · Row-level access
Private pilot · Live collaboration testing pending
If it doesn’t work for us in production, it isn’t finished.
Since September 2026 I’ve worked at DFDSOFT, a DogFoodDev company, as AI Dev Assist. I use Cursor and Codex every day as working partners: to scaffold APIs, explore designs and move faster, while I own the decisions, the review and what ships.
I created BOOM from scratch, a browser recording and collaboration product: browser-native capture with local-first recovery, resumable uploads to private storage, and multi-tenant workspaces with role-based access and PostgreSQL row-level security. I re-architected its media processing from an always-on Dockerized FFmpeg worker to in-browser MP4 encoding with direct upload. I’ve also contributed to Talent Director, building the OAuth 2.0 flow for an MCP connector with dynamic client registration, PKCE, tenant-bound scopes and audit logging, and to Reqev, an applicant tracking platform, adding Microsoft and Google OAuth integrations and GitHub Actions CI with regression, RBAC and HTTP checks.
Created from scratch
Browser recording and collaboration: native capture with local-first recovery, resumable uploads to private storage, and multi-tenant workspaces with RBAC and row-level security. Media processing moved from an always-on FFmpeg worker to in-browser encoding.
TypeScript · Next.js · PostgreSQL · Supabase · Vercel
Contributor
OAuth 2.0 flow for an MCP connector: dynamic client registration, PKCE (S256), tenant-bound scopes and audit logging; CRM REST sync with audit history.
OAuth 2.0 · PKCE · MCP
Contributor
Microsoft and Google OAuth mailbox and calendar integrations, AI/MCP recruiting workflows, and GitHub Actions CI with regression, RBAC and HTTP checks against isolated test databases.
OAuth · GitHub Actions · RBAC
Employer products: no source or internal details are published.
The packet arrives. Then it’s someone else’s turn.
Teaching is how I learned to explain systems clearly. Across Sandip University and the Google developer community, I’ve helped upskill more than 500 engineers in generative AI and cloud.
Next, I want to build cloud and DevOps systems that keep working when things go wrong, with teams who value shipping fast and shipping safely.
Packet delivered.
Open to DevOps and Cloud roles. My CV, GitHub and LinkedIn are one click away.