The problem
Scientific image analysis often means wiring together segmentation, counting, tracking, and reporting tools. The work is slow to assemble, and reproducing a result can depend on manual steps that are easy to lose.
Countify turns a researcher’s plain-language request into an image-processing pipeline that can be inspected and reproduced.
What I’m building
As founder and builder, I own the agent orchestration, vision integration, backend, and product interface. The platform combines planning and execution with specialized scientific tools and researcher validation.
The work spans microscopy use cases including organisms, cells, and bacteria, with annotation and visualization alongside the analysis workflow.
The engineering decisions
Route work to the right tool. MCP orchestration connects planner-executor flows to segmentation, tracking, and z-stack analysis. Specialized vision models and conventional imaging tools have different strengths.
Keep researchers in the loop. Validation gates create explicit opportunities to inspect intermediate results, rather than treating the output of an agent as sufficient evidence.
Support heterogeneous images. The toolset includes vision-language models, SAM, JEPA, RF-DETR, and ImageJ/Fiji. Model selection is tied to the experiment and task.
Build the workflow around the model. A FastAPI backend and Next.js interface support annotation, visualization, and reporting, with CUDA-accelerated inference where appropriate.
A useful scientific assistant needs more than a final answer. The processing steps, intermediate outputs, and human validation are part of the result.
How I approach scientific trust
Per-step validation and a researcher in the loop are core to the architecture. Hybrid vision and VLM-assisted labeling supports dataset curation, with validation gates before labels are accepted.
I treat confidence scoring and reproducible processing as necessary product capabilities. They do not replace domain review or establish universal scientific validity.
Progress and recognition
Countify won first place in the MIA Trek Pitch Competition. It also received an Honorable Mention in the pitch competition at Pakistan Harvard Conference 2026. The platform remains an ongoing founder-led project.
The engineering focus is making a research workflow executable, inspectable, and repeatable across a set of specialized tools.
A public engineering summary. Results are scoped to the project and period described; implementation details are summarized at a high level.