Rojan Dahal
Read in नेपाली

AI Engineer · TitanCloud

Building production ML systems that actually ship.

Most of the work of an LLM pipeline happens upstream of the model. I write about the layers that filter, route, and validate before a single token is spent — and I build them for a living.

Recent Writing

All essays →
8 min

Four layers before the LLM: the gatekeeper pattern

How a stack of cheap classifiers cuts a document-IDP bill by an order of magnitude without ever waking the model.

Read →
6 min

The cheapest token is the one you don't send

Routing, retrieval, and a small set of decisions about which questions a model should never see.

Read →
7 min

Production ML is mostly not ML

A field report from a three-agent IDP pipeline: what breaks first, what's worth automating, and where humans still earn their seat.

Read →

Selected Work

All projects →
2026.03 – present

TitanCloud — Document IDP and the Gatekeeper Pipeline

· AI Engineer

A three-agent IDP pipeline on Amazon Bedrock with a four-layer gatekeeper in front. Filters 92% of input before any LLM token is spent, routes the rest by intent, and closes the loop on human corrections.

  • · Python
  • · AWS Bedrock
  • · EfficientNet-B0
  • · OpenCV
  • · MCP
  • · Neo4j
  • · EventBridge
  • · Amazon A2I
Case study →
2024.09 – 2025.06

Manufacturing surface-defect detection (NAMRC/MSEC 2025)

· Graduate Researcher

A vision pipeline for live surface-defect detection on machined parts, published at NAMRC/MSEC 2025. Combined classical image features with a small CNN to keep inference under 30 ms on workshop hardware.

  • · Python
  • · PyTorch
  • · OpenCV
  • · ONNX Runtime
  • · Jetson Nano
Case study →

About

I’m an AI Engineer at TitanCloud. I hold an MS in Data Science from Gannon University, completed December 2025 with a 4.0 GPA.

Before TitanCloud I shipped on-device computer vision at BitsKraft and ran the shared ML infrastructure for a research group of seven faculty. I care more about the boring layers of a system than the model at the bottom — because the model is rarely what’s broken.

Read the longer version →

Contact

Looking for a full-time AI / ML role starting mid-2026. F-1 OPT, open to relocation.

Best way to reach me is email. I read everything and reply to most things within a day or two.