Avik Mukherjee
Full-stack engineer · security & data tooling
I build full-stack products with a bias toward security and data tooling — and I own what I build, end to end.
Experience
August 2025 - Present
April 2025 - August 2025
Sept 2024 - March 2025
Remote
Trainee Engineer
Nov 2023 - Feb 2024
Remote
Mobile Developer Intern
How I operate
AI gets me to a first draft quickly. Judgement is what makes the draft shippable.
- 01
Start with an agent, not a blank file
I lean on Claude Code and other AI agents to get to a working draft fast — scaffolding, boilerplate, and the first pass at a design. Speed at this stage is cheap, so I spend it freely.
- 02
Then find where it breaks
The draft is a starting point, not an answer. I read what the agent produced, work out the tradeoffs it silently made, and look for the limitations that only show up under real load, real data, or real edge cases. This is where most of the actual engineering happens.
- 03
Talk to the people it affects
Before anything ships, I take the tradeoffs to whoever owns the surrounding system and agree on which ones we're accepting. Then I build the version we actually decided on — not the one that was easiest to generate.
Selected systems
Term Bridge
2026Share a real terminal with anyone using a 6-digit code. No port forwarding, no VPN, no tunnel service.
- 6-digit pairing code
- ·0 bytes of terminal data through my servers
- ·P2P after handshake
MiniQueue
2025Postgres-backed job queue with lease-based semantics and crash recovery. Built to understand the failure modes River, Sidekiq and BullMQ solve.
- at-least-once delivery
- ·35s worst-case crash recovery
- ·30s lease, 10s heartbeat
Distributed Web Crawler
2025Fault-tolerant crawler with URL leasing, per-host politeness, distributed dedup and full-text search across workers.
- 20 concurrent requests per worker
- ·1M-entry bloom filter
- ·300s URL lease TTL
Blog
WebRTC Deep Dive: From Peer to Peer to Global Scale
A full technical breakdown of WebRTC, the protocol behind Zoom, Discord, and Google Meet. Covers why it runs on UDP, the seven protocol layers, NAT traversal with STUN and TURN, the P2P versus SFU scaling problem, and the production gotchas nobody warns you about.
Jun 21, 2026
Why 'AI Will Replace Coders' Is the Wrong Conversation
Everyone's arguing about who's ngmi because of AI. The real question is simpler and a lot less comfortable.
Jun 14, 2026
From 342MB to 37MB: The Go Image I Was Shipping Wrong
I'm new to Go. Our backend image was 342 MB and I had no idea what to do about it. This is the story of learning what each line of the production Dockerfile actually does — multi-stage builds, distroless, CGO_ENABLED=0, and stripping symbols — by getting the size down to 37 MB.
Jun 4, 2026
Development
Sponsor
Support my open-source work
If my projects, blog posts, or tools have helped you, consider sponsoring me on GitHub. Every contribution keeps the side projects shipping.
Contact
Get in touch
Full-time at SuperAlign AI, but always happy to talk security tooling, Go services, data pipelines — or whatever you're building.