Back to work

Agentic systems · Research infrastructure

Tessera

A reproducible agent harness for large-scale content analysis.

Role
Designer & lead engineer
Period
Research → production
The Tessera content-analysis interface
Tessera: recipes, runs, and traceable results.

Tessera lets researchers analyse text corpora far larger than hand-coding allows, without giving up rigour. A YAML recipe DSL defines pipelines, content-addressed result caching makes runs byte-for-byte reproducible, and deferred-event handling makes them fault-tolerant.

The harness is built for audit: prompt hashes, reproducibility manifests, cost and token tracking, schema validation, inter-coder reliability, and post-hoc ablation, exactly the data-integrity guarantees that serious research and policy work demand.