I · 2023
Auctopus Technology
“Automate what you keep checking by hand.”
Frontend Developer Intern
Redux state management and Cypress automation, which took manual testing down by half.
Chapter I. Shadows
Avik Mukherjee, Backend Engineer
We are born facing the happy path. What we call working software is the play of shadows upon it.
Backend Engineer at SuperAlign AI, Hyderabad. Building AI and security infrastructure that has to survive reality.
Scroll down, slowly.
What happens when it drops?
How long did that really take?
What if it arrives twice?
What breaks at ten thousand?
How do you know?
The Wall
Chained to a deadline, we can only look ahead. Behind us runs production, and between it and our backs, requests pass and cast shadows. We name the shadows. We ship on them. We build whole systems upon them.
Shadow I of V
reliable
The network is reliable.
Shadow II of V
instant
Latency is zero.
Shadow III of V
once
Every message arrives once.
Shadow IV of V
later
We can make it scale later.
Shadow V of V
correct
The generated code is correct.
Chapter II. The Turn
Generating code got cheap. Deciding whether it is correct did not.
So someone has to turn around and look at the fire. This is how I do it, drawn as a machine. One crate of records rides the line; scroll to run it.
Step I of V
Anatomy of a workflow
WhereIhaveclimbed.
The way out is steep, and the light blinds. Each place I worked left a way of seeing carved into the rock.
II · 2023 to 2024
Dank
“Slow screens are usually slow requests.”
Mobile Developer Intern
Built most of a React Native app on Expo and cut page load times by 40% through API integration work.
III · 2024 to 2025
DataFoundry AI
“Data has a shape before it has a schema.”
Trainee Engineer
Azure Data Factory pipelines with parameterised queries and Blob Storage, CSV to RDF conversion, and the frontend half of role based access control.
IV · 2025 to now
SuperAlign AI
“Design for the retry, not the request.”
Software Engineer, Backend, now
Backend services, endpoint scanners and ingestion for discovering and governing AI assets across enterprise fleets. Idempotent ingestion, partition aware workers, Postgres locking semantics, and a DuckDB pipeline that chews ~70GB of logs in ~4 minutes.
Nobodyhandedmethe light.I went upfor it,one failureat a time.
Chapter IV. The Light
Outside, I rebuild things until I understand them.
Job queues, a Kubernetes control plane, erasure coding, a Postgres plan visualiser. Then I write down what broke.
Where I look now
- Distributed systems
- Queues, workers, transactional boundaries, idempotency, partitioning, backpressure, failure recovery.
- AI infrastructure
- Asset discovery and inventory, risk assessment, telemetry pipelines, and large volumes of endpoint data.
- Security
- Cross platform scanners, browser extension management, MDM rollout, and risk analysis across macOS, Linux and Windows.
- Data systems
- PostgreSQL, ClickHouse, DuckDB, streaming and object storage. Deciding where data lives and how much work happens synchronously.
What I built to understand it
- Sidecarnew
- MindstateDurable, private memory for AI agents over MCP
- kube-liteA Kubernetes control plane, rebuilt to understand it
- KumoDB
- Term Bridge
- pgxrayPostgreSQL plan visualiser and index advisor
- Vercel Clone
- Paper Improver
- Code Index MCP
- Skills Registry
- AI Gatewayprivate
- Security Scanner
- bridgecord
- Plane Discord Bot
Epilogue
The engineer who has seen production goes back down to the planning meeting. They laugh: the happy path is faster, they say. Better to ship the shadows.
No system leaves the cave once and for all. Engineering is the climb, and it has to be made again on every pull request, by anyone willing to ask what happens when it breaks.
Working on hard problems in infrastructure, AI systems, security or developer tooling? I'd like to hear about it.