mohanish3/eval-atlas

How mohanish3/eval-atlas works

Eval Atlas is a browser-based LLM evaluation dashboard. Users create or upload an eval set, configure a run and model providers, then inspect streamed question results and category-level accuracy. A second workflow runs prompt research: select a target model and research model, iterate and score candidate prompts, inspect trials, and promote the best prompt. The tree and excerpts establish the research route’s orchestration and SSE handling; several main-evaluation and persistence connections are inferred only at the subsystem level, so uncertain internal calls are omitted.

Main entry points: promptResearch.ts Research routes · App.tsx App routes · EvalResearch.tsx Research view · server.ts Express server

Components 22

Dashboard

App routes frontend/src/App.tsx
Eval pages frontend/src/pages
Research view frontend/src/pages/EvalResearch.tsx
Run stream hook frontend/src/hooks/useEvalStream.ts
Research stream hook frontend/src/hooks/useResearchStream.ts
Run state frontend/src/store/useEvalStore.ts
API client frontend/src/services/api.ts

API

Express server backend/src/server.ts
Eval routes backend/src/routes/evals.ts
Research routes backend/src/routes/promptResearch.ts

Evaluation

Eval sets backend/src/routes/evals.ts
Eval runner backend/src/evals/runner.ts
Answer scorer backend/src/evals/scorer.ts
Model adapters backend/src/evals/adapters

Prompt research

Research loop backend/src/evals/promptResearchRunner.ts
Research budget backend/src/evals/researchBudget.ts
SSE registry backend/src/evals/sseManager.ts

Persistence

Eval fallback store backend/src/evals/fallbackStore.ts
Research store backend/src/evals/promptResearchStore.ts
PostgreSQL backend/src/db/connection.ts

Other components

Browser user
LLM providers

Connections 24

Browser user → App routes: uses
App routes → Eval pages: routes to
App routes → Research view: routes to
Eval pages → API client: requests data
Research view → API client: loads and promotes
Research view → Research stream hook: subscribes
Run stream hook → Run state: updates results
Research stream hook → SSE registry: receives events
API client → Express server: HTTP requests
Express server → Eval routes: mounts routes
Express server → Research routes: mounts routes
Eval routes → Eval runner: runs evaluations
Eval runner → Answer scorer: scores answers
Eval runner → Model adapters: queries models
Model adapters → LLM providers: sends prompts
Research routes → Eval fallback store: loads memory set
Research routes → PostgreSQL: loads eval set
Research routes → Research budget: resolves budget
Research routes → Research store: creates and reads runs
Research routes → Research loop: starts asynchronously
Research routes → SSE registry: registers stream clients
Research loop → Research store: persists research
Research loop → SSE registry: broadcasts trials
Research store → PostgreSQL: persists runs

More TypeScript diagrams: delsner/python-flask-angular-starter · dodopayments/dodo-supabase-subscription-starter · google-marketing-solutions/giga · holymaiden/wedding-app · jolocom/jolocom-sdk

Drawn by GitDiagram on September 26, 2026 from the repository's file tree, README and source files. Read as Markdown ·