Evolv
Prompt evolution platform that treats prompts like genetic sequences. It extracts a DNA fingerprint from every prompt, clusters them semantically with embeddings, synthesizes templates from the clusters, and tracks each mutation and its lineage. Won 1st prize at the Portkey AI Builder Challenge.
Problem
Prompt sprawl: teams accumulate hundreds of near-duplicate prompts with no lineage, no versioning, and no way to know which mutation of a prompt actually improved it. Improvements get lost; regressions ship silently.
Approach
- Treat prompts like genetic sequences: extract a 'DNA' fingerprint automatically from any prompt.
- Cluster prompts semantically with embeddings so variants of the same intent surface together.
- Synthesize templates from clusters and track mutations plus lineage across versions.
- Turn the whole thing into a product: search, compare, and promote templates instead of copy-pasting prompts.
Outcome
A platform that converts prompt sprawl into organized, versioned templates with full lineage. Won 1st prize at the Portkey AI Builder Challenge.
Architecture
dna = extract_dna(prompt) # structure + intent
cluster = embed_and_cluster(dna)
template = synthesize(cluster)
child = template.mutate(edit)
lineage.record(parent=template.id,
child=child.id)Highlights
Automatic prompt 'DNA' extraction from existing prompts.
Semantic clustering with embeddings.
Template synthesis with mutation and lineage tracking.
Turns prompt sprawl into organized, versioned templates.
1st prize at the Portkey AI Builder Challenge.
Stack
- Python
- Next.js
- TypeScript
- PostgreSQL
- OpenAI API