Skip to content
BranchesTags

Repository files navigation

AllCallAll

AllCallAll is a full-stack engineering portfolio project for an enterprise realtime collaboration and AI-assisted management system. It is built around React/Vite, Go, Gin, Gorm/MySQL, Redis, WebSocket/WebRTC, recording storage, asynchronous workers, Elasticsearch search, and AI Agent workflows.

The strongest project story is not a commercial app launch. The repo demonstrates how an enterprise collaboration product can evolve from a modular monolith into a maintainable full-stack system with generated API contracts, an admin dashboard, realtime collaboration, performance gates, extracted worker/service boundaries, a primary Web app, native Expo mobile app, and Electron shell.

Current Positioning

  • Realtime collaboration: organizations, conversations, messages, notes, durable WebSocket replay, room state, WebRTC signaling, and recording sessions.
  • Enterprise admin console: organization overview, member/invite/team/policy/audit management, admin summary API, Redis-backed dashboard cache, and reproducible dashboard/message-list benchmarks.
  • AI Agent system: ReAct-style single-agent runs, workflow/DAG-style multi-agent tasks, tool calling, approvals, persisted traces, memory, and conversation-aware context retrieval. Beta/demo Compose defaults to a Python FastAPI + LangGraph Agent Runtime while Go remains the source of truth for data, permissions, approvals, audit, and writes.
  • Meeting recording transcription: recording stop can enqueue recording.transcription.requested; the provider abstraction supports mock and OpenAI-compatible ASR paths and stores MeetingTranscriptSegment rows for Agent retrieval.
  • Hybrid retrieval: conversation context plus external knowledge sources, BM25/vector search where Elasticsearch is configured, Go-layer reranking/chunk assembly, and an optional Python RAG Runtime for Agentic retrieval orchestration, rerank, evidence packs, and grounding checks.
  • Async reliability: MySQL outbox, claim/lease workers, retries, idempotency keys, and Prometheus-style counters.
  • Extractable runtime: API server can run embedded workers locally, or split into User Service, Agent Worker, Outbox Worker, Data Worker, Search Worker, and Cleanup Worker.
  • Primary Web surface: web/ is the main browser client, covering auth, organizations, collaboration, chat, calls, meetings, recordings, transcripts, Agent Lab, knowledge, approvals, billing settings, Web Push, and responsive layouts.
  • Supporting native/desktop surfaces: mobile/ remains the Expo Android/iOS client; desktop/ is a thin Electron shell over the Web client.
  • Beta v1 scope: the practical product loop is a 3-6 person team flow: create organization, invite members, chat, start a meeting, record, transcribe, generate an Agent meeting brief, inspect citations, and approve write-back tools.

Realtime translation code is still present for compatibility, but the mobile UI entry points are currently hidden. Meeting transcription is now independent of the realtime translation switch.

Repository Map

backend/     Go backend: API, auth, collaboration, Agent adapters, search, storage, workers
web/         Primary React + Vite + TypeScript Web application
mobile/      Expo React Native app for native Android/iOS
desktop/     Electron shell wrapping the Web client
infra/       Docker Compose local stack and optional interview infra profiles
scripts/     Development, smoke, seed, and benchmark scripts
docs/        Current docs, interview docs, deployment notes, and selected references

The Python Agent/RAG runtime source has been split into XianingY/allcallall-agent-runtime. AllCallAll integrates it over HTTP and Docker images; Go remains the source of truth for product data, permissions, approvals, audit, and write execution.

Fast Start

Start MySQL and Redis:

./scripts/development/start-services.sh

Run the backend API:

cd backend
CONFIG_PATH=./configs/config.yaml go run ./cmd/server

Health check:

curl http://localhost:8080/api/v1/health

Run the Web app:

cd web
npm install
npm run dev

Vite serves the app at http://localhost:5173 and proxies /api and WebSocket traffic to http://127.0.0.1:8080.

Run the desktop shell:

cd desktop
npm install
ALLCALLALL_WEB_URL=http://localhost:5173 npm run build

Android development still uses Expo/React Native and EXPO_PUBLIC_* configuration. APP_ENV is historical and is not an active mobile runtime control.

Backend Commands

# API server, embedded workers enabled by default
make run-api

# Extracted service/process boundaries
make run-user-service
make run-agent-worker
make run-outbox-worker
make run-data-worker
make run-search-worker
make run-cleanup-worker
make run-agent-runtime
make run-rag-runtime

# Deterministic interview/demo harnesses
make beta-seed
make interview-demo
make interview-smoke
make interview-chaos
make interview-down
make interview-microservice-demo
make agent-eval
make rag-eval
make rerank-eval
make python-agent-eval
make python-rag-eval
make ai-agent-jd-eval
make ai-portfolio-eval
make dashboard-bench
make realtime-replay-bench
make chat-ws-replay-bench
make web-contract-check
make web-performance-check

make run-backend is retained as a compatibility alias for the API server.

Runtime Configuration

The main API server, Agent Worker, and MCP tool server call godotenv.Load() as a development convenience, then load CONFIG_PATH YAML and environment overrides. In production or Docker Compose, inject variables explicitly.

Common backend variables:

  • CONFIG_PATH: backend YAML config path.
  • DB_DSN: MySQL DSN override.
  • REDIS_ADDR, REDIS_PASSWORD: Redis connection.
  • JWT_SECRET: JWT signing secret.
  • MAIL_PASSWORD: SMTP credential override.
  • AGENT_PROVIDER=rules|mock_llm|openai_compatible: Agent planner/provider selection.
  • AGENT_PROVIDER_STRICT=true: required for Beta/production when using a real provider; disables silent fallback to rules.
  • RAG_RERANK_ENABLED=true, RAG_RERANK_PROVIDER=rules|cross_encoder_compatible: optional post-retrieval rerank over BM25/vector/RRF candidates.
  • AGENT_RUNTIME=python_langgraph|legacy_go: Agent orchestration runtime. Compose/Beta demo defaults to python_langgraph; bare Go processes without this env still use the legacy in-process runtime. go is accepted as a legacy alias.
  • PY_AGENT_RUNTIME_BASE_URL: Python LangGraph runtime URL, defaulting to http://127.0.0.1:8090 locally and http://agent-runtime:8090 in Compose.
  • PY_RAG_RUNTIME_BASE_URL: Python RAG Runtime URL, defaulting to http://rag-runtime:8091 in Compose when Agentic RAG calls are enabled.
  • PY_AGENT_RUNTIME_IMAGE, PY_RAG_RUNTIME_IMAGE: Docker image overrides. Compose defaults to ghcr.io/xianingy/allcallall-agent-runtime/{agent-runtime,rag-runtime}:v0.2.0.
  • PY_AGENT_RUNTIME_TIMEOUT_SEC, PY_AGENT_RUNTIME_STRICT: timeout and strict failure behavior for the Python runtime.
  • AGENT_RUNTIME_TOOL_TOKEN: shared token that protects the Go read-only tool bridge for Python runtime calls.
  • PY_AGENT_PROVIDER=rules|openai_compatible: Python runtime provider selection.
  • PY_AGENT_OPENAI_BASE_URL, PY_AGENT_OPENAI_MODEL, PY_AGENT_OPENAI_API_KEY: Python runtime OpenAI-compatible chat provider settings.
  • PY_AGENT_PROMPT_VERSION, PY_AGENT_ENABLE_GROUNDING_CHECK: Python prompt-template and grounding-check controls.
  • PY_AGENT_ENABLE_AGENTIC_RAG=false, PY_AGENT_RAG_MAX_RETRIEVAL_STEPS=3, PY_AGENT_RAG_MIN_CONFIDENCE=0.6: bounded Agentic RAG planning/refinement controls.
  • PY_AGENT_TOOL_BRIDGE_BASE_URL, PY_AGENT_TOOL_BRIDGE_TOKEN: Python runtime access to Go-owned read-only tools.
  • PY_RAG_TOOL_BRIDGE_BASE_URL, PY_RAG_TOOL_BRIDGE_TOKEN: Python RAG Runtime access to the Go internal authorized retrieval bridge.
  • PY_RAG_RERANK_PROVIDER=rules|cross_encoder_compatible, PY_RAG_TOP_K, PY_RAG_MAX_STEPS, PY_RAG_MIN_CONFIDENCE: Python RAG Runtime retrieval and rerank controls.
  • PY_AGENT_HARNESS=allcallall_v1, PY_AGENT_LOOP_MAX_STEPS=5, PY_AGENT_ENABLE_CRITIC=true: Python Agent Runtime Harness and bounded-loop controls.
  • PY_RAG_VECTOR_STORE=none|qdrant, PY_RAG_QDRANT_URL, PY_RAG_QDRANT_COLLECTION: optional Qdrant adapter for vector/payload retrieval experiments; Go remains the production data boundary.
  • EMBEDDED_WORKERS=0|1: controls API-process workers.
  • FCM_SERVICE_ACCOUNT_PATH: enables Firebase Admin SDK push delivery.
  • RECORDING_STORAGE_DRIVER=local|s3: recording storage driver.
  • RECORDING_STORAGE_DIR: local recording storage root.
  • RECORDING_S3_*: S3-compatible recording storage config.
  • TRANSCRIPTION_ENABLED=true: enables recording transcription jobs.
  • TRANSCRIPTION_PROVIDER=mock|openai_compatible: recording transcription provider.
  • TRANSCRIPTION_OPENAI_BASE_URL, TRANSCRIPTION_OPENAI_MODEL, TRANSCRIPTION_OPENAI_API_KEY: OpenAI-compatible ASR settings for Beta.
  • USER_SERVICE_GRPC_ADDR: enables API auth validation through User Service.
  • KAFKA_BROKERS, KAFKA_SETTLEMENT_TOPIC: enables settlement event publishing/consumption.
  • ELASTICSEARCH_URL, ELASTICSEARCH_INDEX: enables Elasticsearch message/search indexing.

Web runtime public variables are injected through /config.js in production:

  • PUBLIC_API_BASE_URL
  • PUBLIC_WS_BASE_URL
  • FIREBASE_API_KEY, FIREBASE_AUTH_DOMAIN, FIREBASE_PROJECT_ID, FIREBASE_STORAGE_BUCKET, FIREBASE_MESSAGING_SENDER_ID, FIREBASE_APP_ID, FIREBASE_VAPID_KEY
  • REVENUECAT_PUBLIC_API_KEY

Mobile native public variables:

  • EXPO_PUBLIC_API_HTTP
  • EXPO_PUBLIC_API_WS
  • EXPO_PUBLIC_FORCE_TLS
  • EXPO_PUBLIC_RESTRICTED_NETWORK
  • EXPO_PUBLIC_SIGNALING_TRANSPORT
  • EXPO_PUBLIC_SIGNALING_SHAPING
  • EXPO_PUBLIC_E2EE_MODE=experimental
  • EXPO_PUBLIC_REVENUECAT_API_KEY and related RevenueCat variables for Android subscription paths.

Optional Microservice / Infra Demo

Run local infrastructure for the interview-oriented extracted topology:

docker compose -f infra/docker-compose.yml \
  --profile microservices \
  --profile interview-infra \
  up api user-service outbox-worker data-worker search-worker kafka elasticsearch

Important variables for that topology:

USER_GRPC_ADDR=:9090
USER_SERVICE_GRPC_ADDR=user-service:9090
KAFKA_BROKERS=kafka:9092
KAFKA_SETTLEMENT_TOPIC=allcallall.room.settlements
ELASTICSEARCH_URL=http://elasticsearch:9200
ELASTICSEARCH_INDEX=allcallall_messages
EMBEDDED_WORKERS=0

Beta Demo Seed

For local Beta walkthroughs, seed a small team workspace after MySQL is running:

make beta-seed

The command creates deterministic owner/member accounts, an organization, a team, a conversation, sample messages, a completed meeting, ready transcript segments, and audit events. It prints login credentials and direct Web routes. The seeded transcript is metadata and text-only; it is useful for UI and Agent grounding demos, not for real ASR/download validation.

Documentation

Start with:

Quality Gates

Backend:

cd backend && go test ./... && go vet ./...

Web:

cd web && npm run typecheck && npm run lint && npm test && npm run build
cd web && npm run contract:check && npm run bundle:budget

Mobile:

cd mobile && npx tsc --noEmit && npm run lint

Desktop:

cd desktop && npm run check && npm run build

Use the narrowest relevant check for small changes, then broaden when shared behavior or generated surfaces are touched.

Scope Boundaries

  • Kubernetes is intentionally not part of the current implementation.
  • Transcript editing, large-meeting SFU scale-out, and Kubernetes are future enhancements.
  • Kafka and Elasticsearch adapters exist, but live smoke tests require the optional Compose profiles.
  • Production Web billing and Web Push require RevenueCat/Firebase public runtime config plus backend FCM service-account config.
  • The .proto file is the source contract; current Go gRPC binding is hand-written because protoc is not required by CI.

License

This repository is used as a personal engineering portfolio project. Confirm licensing requirements before commercial reuse.

About

Final Year Project

Resources

Readme

Contributing

Contributing

Security policy

Security policy
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages