Backend Engineer — Node.js · NestJS · TypeScript · Python
I build backends that survive real load and real change: secure by default, tested before merge, observable after deploy.
Open to backend roles where engineering quality is non-negotiable — Email · LinkedIn
role: Backend Engineer
focus: secure APIs · multi-service architecture · quality gates
current: leading backend for Afaq (NestJS + FastAPI)
shipped: SemCodec (open source) · LinkHub (desktop) · client backendsSending motion, not pixels
Problem: video calls collapse on weak networks — codecs keep sending full pixel frames even when most of the frame hasn't changed.
Solution: a TypeScript middleware that extracts facial landmarks on the sender, transmits only motion parameters (~1–3 KB/frame, zero pixel data), and reconstructs video client-side via MediaPipe + WebAssembly. No server GPU required.
Result: open-source prototype with an estimated ~70% bandwidth reduction from skipped pixel frames. Full quality validation and real-network testing are still in progress.
TypeScript · MediaPipe · WebAssembly · HTML5 Canvas · MIT
One workflow for all your Android tools
Problem: managing Android devices from a computer is fragmented — file transfer, screen mirroring, and media downloads each require a separate tool and workflow.
Solution: a cross-platform Electron/Node.js app that connects over ADB, mirrors the screen with low latency, transfers files, and runs a yt-dlp media engine (1000+ sites) that pushes content to connected devices automatically.
The hard part: each CLI tool (ADB, scrcpy, yt-dlp) has its own output format and failure modes. I built a process-management layer from scratch with a dedicated wrapper per tool — normalizing output and isolating errors from the orchestration logic.
Result: shipped as a published desktop app, validated on real hardware. Led a 5-member team and implemented the core system end-to-end.
Electron · Node.js · ADB · scrcpy · yt-dlp
Private repo — architecture summarized below
Problem: scholarship discovery is fragmented across dozens of sources; applicants get no end-to-end guidance.
Solution: NestJS core backend that aggregates opportunities, matches them to user profiles, and supports the full application lifecycle (including an AI-assisted document editor). Coordinates a two-developer Python/FastAPI service for semantic matching, scraping, and data cleaning.
Key decisions:
- Shared database + modular monolith over microservices — cut operational cost at the current stage while keeping the Python service isolated behind NestJS for security and latency.
- Moved the AI assistant into NestJS — removed a network hop and kept the Python service fully behind the API surface.
- Delivered the AI document editor in 2 days by integrating a compatible open-source project instead of building from scratch.
- Quality pipeline before merge: ~80% test coverage (unit/integration/E2E), SonarQube, ArchUnitTS, k6, Trivy — every change validated.
NestJS · FastAPI · PostgreSQL · Redis · Docker · GitHub Actions
Detailed engineering write-ups live in the repo (private). Happy to walk through them on request.
01 — Design for the trade-off, not the trend. I chose a modular monolith over microservices on Afaq because the complexity cost outweighed the scaling benefit at that stage. Architecture is a decision, not a decoration.
02 — Quality is infrastructure, not a phase. Every change passes a pipeline — tests, static analysis, architectural enforcement, load testing, dependency scanning — before merge, not after the bug ships.
03 — Go below the framework when the framework isn't enough. For LinkHub, process control, IPC, and tool failures needed their own layer. Frameworks solve framework problems.
| Area | What I actually do |
|---|---|
| Backend architecture | Modular service design, shared-database trade-offs, service isolation behind one API surface |
| API design & security | REST, RBAC, input validation, structured error handling, request tracing |
| Quality engineering | CI/CD, unit/integration/E2E testing, static analysis, arch enforcement, load testing, dependency scanning |
| Process & tool orchestration | OS-level process management, CLI wrappers, output normalization, failure isolation |
| AI integration | LLM agents with tool use, RAG, embeddings, semantic search |

