AI backend engineer
I build the backend around the model: agents that call tools on live systems, semantic search on pgvector, and self-hosted inference, with the same failure handling as any other production service.
Practicalities
- Based in
- Lahore, Pakistan (PKT (UTC+5))
- Experience
- 3+ years, full-time
- Now
- Lead Software Engineer, TransData
Evidence
Three builds that match this role.
-
AI agents working inside a live ERP
Built an agent platform over an ERP: forced tool-calling, one versioned contract, multi-company resolution.
-
A harness for AI coding agents near production
Ran AI coding agents next to production with 11 project skills, 3 agent definitions and human-approved fixes.
-
A social and market signal intelligence backend
Built ingestion, AI summaries, pgvector search and alerting alone for a client.
Stack for this role
Questions
Has he run language models himself, or only called APIs?
Both. He set up self-hosted inference with vLLM behind an OpenAI-compatible LiteLLM proxy on one 12 GB GPU, and the engineering team adopted it.
Does he train models?
His production AI work is the backend around models: agents, tool-calling, search and inference. His model training was research, in a peer-reviewed ECG paper (Springer, 2025).
How does he use AI coding tools?
As a supervised team: read-only analysis agents in parallel, a stronger reviewer model on decisions, and a human approving every fix, with rules that trace to real incidents.