Askbase

Your support team answers questions in seconds, not hours.

Askbase is a RAG-powered knowledge retrieval platform for support teams. Ingest your docs, ask questions in natural language, get grounded answers with precise citations — no hallucinations, no guesswork.

---

Before & After

	Without Askbase	With Askbase
Find an answer	Search 15 docs, scan 200 pages, hope you found the right one	Ask a question, get the answer with source links in <100ms
Onboard new agents	2-week ramp-up reading tribal knowledge	Day-one accuracy: the system knows what the docs say
Audit trail	"I think I read it somewhere..."	Every answer cites doc name, section, and character offsets
Consistency	Different agent, different answer	Same question = same grounded answer every time

---

Sample Q&A

POST /ask
{"query": "How do I reset my password?", "top_k": 3}

{
  "answer": "Answer grounded in retrieved context:\n[1] Go to Settings > Security > Reset Password. You will receive a confirmation email within 2 minutes...",
  "citations": [
    {
      "doc_id": "faq.md",
      "start_char": 108,
      "end_char": 328,
      "score": 0.7234,
      "excerpt": "Go to Settings > Security > Reset Password. You will receive a confirmation email within 2 minutes..."
    }
  ],
  "citations_count": 1
}

Every answer traces back to the source. No black boxes.

---

Try It in 60 Seconds

Option A: Docker (zero setup)

git clone https://github.com/dsantoreis/askbase.git && cd askbase
docker-compose up
# API is live at http://localhost:8080 with sample docs pre-loaded

Option B: Local

git clone https://github.com/dsantoreis/askbase.git && cd askbase
python3 -m venv .venv && source .venv/bin/activate
pip install -e '.[dev]'
rag serve --seed --host 0.0.0.0 --port 8080
# --seed auto-ingests sample FAQ, product guide, and troubleshooting docs

Query it

curl -s -X POST http://localhost:8080/ask \
  -H "Authorization: Bearer user-demo-token" \
  -H "Content-Type: application/json" \
  -d '{"query": "What plans are available?"}' | python3 -m json.tool

See DEMO.md for a full walkthrough with real queries and answers.

---

Platform Architecture

┌─────────────────────────────────────────────────┐
│  Frontend (Next.js 14)                          │
│  /chat — end-user query interface               │
│  /admin — operational dashboard                 │
└──────────────┬──────────────────────────────────┘
               │ REST API
┌──────────────▼──────────────────────────────────┐
│  Backend (FastAPI)                               │
│  POST /ingest    — admin: index documents        │
│  POST /ask       — user: query knowledge base    │
│  POST /evaluate  — admin: measure retrieval      │
│  GET  /metrics   — Prometheus-compatible          │
│  GET  /health    — liveness probe                │
│  GET  /readyz    — readiness probe               │
│  GET  /admin/stats — operational statistics      │
├──────────────────────────────────────────────────┤
│  Pipeline: Documents → Chunks → TF-IDF Index     │
│  Auth: Bearer tokens (admin/user roles)          │
│  Rate limiting: per-client, 30 req/min           │
└──────────────────────────────────────────────────┘

---

Integration Points

Askbase is API-first. Plug it into your existing workflow:

• Slack: Use the /askbase command or Slack bot to query from any channel
• Microsoft Teams: @mention the Askbase bot for inline answers with citations
• REST API: Full CRUD via Bearer-token authenticated endpoints
• Webhooks: Get notified on ingestion, new answers, or evaluation results
• Custom frontends: The API returns structured JSON — build any UI on top

---

Performance

Metric	Value
p50 latency	42ms
p95 latency	91ms
Chaos recovery	6s
Soak test	5 min / 0% errors

Benchmarked locally with TF-IDF baseline. See results/perf-results.md for full report.

---

Security

• Bearer-token auth with admin/user role separation
• Per-client rate limiting (429 on overflow)
• Structured audit logging on every request
• Prometheus metrics for monitoring
• SOC 2 Type II architecture patterns

Configure tokens via environment variables:

export RAG_ADMIN_TOKEN=your-admin-token
export RAG_USER_TOKEN=your-user-token

See .env.example for all configuration options.

---

Test Strategy

• Unit: chunking boundaries, retrieval scoring, pipeline internals
• Integration: full ingest → ask → evaluate flow
• API: auth enforcement, rate limiting, error handling
• E2E: frontend navigation with Playwright
• Coverage gate: --cov-fail-under=80 enforced in CI

pip install -e '.[dev]'
pytest --cov=rag_pipeline --cov-report=term-missing

---

Deployment

Target	Guide
Docker	docker-compose up
AWS	docs/deploy/aws.md
GCP	docs/deploy/gcp.md
Bare metal	docs/deploy/bare-metal.md
Kubernetes	Manifests in k8s/ — namespace, deployments, services, ingress

---

Project Structure

askbase/
├── src/rag_pipeline/    # FastAPI backend + RAG engine
├── frontend/            # Next.js 14 chat + admin UI
├── data/seed/           # Sample docs for demo mode
├── k8s/                 # Kubernetes manifests
├── scripts/             # Load, chaos, and soak tests
├── tests/               # Unit, integration, API tests
├── docs/                # Architecture + deployment guides
└── docker-compose.yml   # One-command local deployment

---

Roadmap

• Semantic embedding support (sentence-transformers, OpenAI ada)
• Multi-tenant document isolation
• Streaming answers via SSE
• Document change detection and auto-reindex
• Slack/Teams bot as first-class integration

---

License

MIT -- see LICENSE.