A conversation needs a next step
Product information, business rules and order details need to stay connected. Tanet explores this in conversational commerce for Bangla and Banglish shop interactions.
PRACTICE / AI SYSTEMS
Intelligence needs a useful structure.
A model is one part of the system. The harder questions are what context it receives, what it may change, how it hands work back, and how someone can tell whether it helped.
The input is larger than the context
Video contains more evidence than an agent can inspect at once. Morphena explores selective, time-referenced representations that preserve a route back to the source.
The workflow needs boundaries
The useful unit is the complete handoff: input, constraints, processing, review and an observable result. Scope and evaluation belong in the first conversation.
How I approach
the work.
01 / Understand the task and its failure modes
02 / Choose a bounded workflow and real inputs
03 / Build the integration and review path
04 / Evaluate the output against the intended use
RELATED WORK / THE DETAILS ARE OPEN
LIVE SITE / 01 ↗
Tanet
↗From a shop conversation to an actionable order.
Product work · Kronx
SOURCE / 02 ↗
Morphena
↗Let an editing agent inspect a film without swallowing every frame.
Open-source toolmaking
LIVE SITE / 03 ↗
Elite Heights
↗The systems behind a digital practice.
Systems & AI engineering
Start with
your problem.
Share the context, the people who need it, and the part that is difficult. We can work out what is worth building.