portfolio-synapse
Portfolio narrative for Synapse, framed as "semantic search and relationship-mapping engine over personal networks." Status: Live.
Key extractions
- Thesis: turn unstructured social graph into a queryable knowledge base; NL queries like "who do I know that's built fintech products?"
- Stack: OpenAI embeddings, Ollama for local query parsing, Next.js frontend, Rapid API for contact enrichment.
- Key decision: composite scoring (semantic similarity + interaction recency + relationship strength) over pure cosine — naive semantic search returns too many false positives.
- Key decision: Ollama-local query parsing before embedding search to extract intent/constraints and avoid the "everything is similar" problem.
- Cold start problem: sparse contact data (just name+email) isn't enough for a useful embedding.
- Constraint: privacy — every design decision had to account for data minimization and user control.
Related
- Cross-references:
orqysexplicitly in "Related" section of the portfolio page.
Tension flag
The github repo named synapse describes a Twitter follower analysis tool, not a contact semantic-search engine. Either the portfolio is forward-looking / aspirational, or the project pivoted, or these are unrelated builds under the same name. Captured in synapse project page.