shernan javier

building systems that turn messy, live signals into structured knowledge.
infra, abstraction, and browser tools for humans + agents to reason over.

or contact me directly

projects

things i've shipped

product-first builds backed by technical primitives: crawling, browser runtime hooks, event pipelines, and explainable scoring.

creator economy · partnership intelligence

creatorgraph

brand-side dealflow engine for stan-style ecosystems. converts a brand url into a structured dossier, generates campaign briefs, ranks creators with explainable scoring, and outputs outreach grounded in real signals.

brand ingestion and dossier generation from web signals
deterministic scoring and ranking with stored match edges
brief and outreach generation grounded in brand and creator features

normalized creator and brand features into a scoring-friendly schema for explainability
designed a scoring breakdown so brands can audit why a creator ranked
iterating toward richer signals such as topic embeddings and audience fit priors

matchingpipelinesrankingagentspostgres

view repo

wallet safety

ensight

browser extension + backend that sits in the path of ethereum wallet requests and turns them into structured intent events, so users get readable explanations instead of opaque hex.

runtime interception of wallet request flows and transaction intents
structured intent event schema for downstream analysis and UI
lightweight risk heuristics and explainable breakdowns pre-signing

focused on turning wallet calls into a consistent event contract across providers
optimized for fast local analysis so the user experience stays interactive
designed for traceability: from raw call to parsed intent to explanation

browser runtimeinterceptionevent schemasweb3

view repo

documentation infra

docforge

documentation compiler and shared docset store for dynamic, js-heavy docs. renders pages, normalizes content, versions artifacts, and exposes fast search plus an agent-facing api for retrieval.

playwright rendering for js-heavy docs and consistent extraction
versioned doc artifacts with freshness checks and diffs
agent-facing retrieval surfaces to reduce repeated crawling

separates render, extract, normalize, and index into explicit stages
designed cacheable artifacts so multiple consumers reuse the same docsets
building toward structured knowledge outputs with stable ids over time

playwrightqueuespostgressearchagent api

view repo

community graph · consent-first intros

cansbridge connects

internal connection layer for cansbridge scholars and fellows designed to reduce social pressure. sender requests an intro, recipient privately accepts or ignores, and only approvals create a connection.

recipient-gated intros that remove obligation and awkwardness
preference-aware connection intent to separate friends vs hiring asks
graph densification for senior retention and community continuity

designed the intro flow so the recipient holds the power by default
optimizing for retention and repeated engagement, not hiring throughput first
future phase: discovery layer that increases surface area without spam

workflow designcommunitygraphconsent-first

view

personal data systems

daytrace

personal data system for structured life signals like location, habits, and activity. maps streams into an ontology so agents and humans can query an explainable graph of daily state.

stream to ontology mapping for durable, queryable state
traceable primitives for reflection and agent workflows
data modeling under uncertainty with explainability first

focus on stable concepts and ids so longitudinal analysis stays coherent
prioritizes readable representations over black-box summaries
built as a foundation for future agent tooling over personal data

ontologiesgraphpersonal analyticsdata modeling

view repo

browser telemetry

webtrace

browser sdk for live interaction and performance signals. captures fine-grained client events and streams structured data into a pipeline for analytics and optimization.

client-side instrumentation with structured event schemas
low overhead capture for behavior and timing signals
stream-friendly payloads designed for real-time analysis

built around contracts that stay stable as events evolve
optimized for signal quality without turning into surveillance bloat
designed for replayability and debugging of event flows

browser sdktelemetryevent schemasanalytics

labs & experiments

experiments

small, focused builds exploring system design, browser internals, and agent-native interfaces. each one makes the reasoning and tradeoffs explicit instead of hiding them.

system design sketches and working prototypes
browser internals exploration and tooling
agent-native interfaces and docs experiments

kept as a public engineering notebook of constraints and decisions
used to test primitives before promoting them to larger products

system designbrowser internalsagents

view repo

experience

early systems builder

founding-engineer style work across browser tooling, pipelines, agents, and marketplaces. end-to-end ownership from primitives and abstractions to hardening under real traffic.

founding engineer

stealth startup

shipped a b2b behavioral intelligence platform with a browser sdk, streaming pipelines, and inference layers for adaptive marketing systems.

browser sdk instrumentation
streaming / event pipelines
inference / optimization layer

designed the event schema and contracts between client, stream, and inference layer
made tradeoffs around latency vs depth of analysis for in-session decisions
built internal tooling to replay traffic and validate new inference logic

browser sdkpipelinesinferenceb2b

founding engineer

clover labs

built ai-driven growth agents and early infrastructure for scalable distribution systems.

agent workflows and orchestration
automation pipelines
early platform infra

defined the core agent loop: perception → planning → action → logging
built guardrails and failure modes to keep agents debuggable, not magic
set up basic observability around agent runs and downstream effects

agentsautomationinfra

co-founder & cto

merchme

led product architecture and full-stack development for a ugc marketplace connecting creators and brands; supported 100+ creators and active brand partners.

marketplace core and matching
campaign / bounty flows
ops and admin tooling

designed data model for creators, brands, campaigns, and payouts
built internal dashboards for campaign health, creator performance, and fulfillment
iterated quickly on flows with tight feedback loops from brands and creators

marketplacefull-stackproduct

engineering internships (selected)