Podifact Docs

Dashboard sections

Pipeline overview

JSON format (structured)

Preferred format. Enables speaker attribution, timestamps, and per-section chunking.

{
  "episode_id": "ep001",
  "title":      "AI in Healthcare: Hype or Reality?",
  "host":       "Alex Chen",
  "guests":     ["Dr. Priya Sharma"],
  "transcript": [
    {
      "timestamp": "00:00",
      "speaker":   "Alex Chen",
      "section":   "Introduction",
      "text":      "Welcome back to the show..."
    }
  ]
}

Plain text format

TXT files are accepted. Speaker detection and sectioning are inferred by the LLM. Less accurate than JSON.

Processing options

Library tab

Episode card actions

Semantic Search tab

Search by meaning across all indexed episodes. Results are ranked by cosine similarity and include the matching transcript chunk, episode reference, and section label. Useful for finding specific claims or topics across your entire library.

Claim statuses

Confidence score

A 0–100% score representing the LLM's certainty in its verdict given the available evidence. High confidence (≥75%) with verified status = strong result. Low confidence with verified = treat with caution.

Source basis

Starting a conversation

Click + New chat in the sidebar, or navigate from an episode card using the chat icon — the latter auto-creates a conversation scoped to that episode.

Suggested prompts

When starting a scoped conversation, prompt suggestions are generated from the episode title (e.g. "What were the key verified claims in…"). For general chats, suggestions cover your full library.

Conversation management

Supported formats

Processing lifecycle

Job statuses

Stats columns

Profile

Update username and email independently. Username must be 3+ characters, alphanumeric with underscores/hyphens. Both fields are pre-populated from your account on load.

Password change

Requires your current password. New password must be 8+ characters. Separate from the profile save — uses its own submit button.

API Keys

Generate long-lived personal API keys for use in scripts and integrations. Keys are prefixed pa_live_ and grant the same access as your session token.

JWT tokens (short-lived, 30 min)

curl -X POST http://localhost:8000/api/v1/auth/login \
  -H "Content-Type: application/json" \
  -d '{"username":"alice","password":"password123"}'

API keys (long-lived, pa_live_...)

Generate personal API keys in Settings → API Keys. Use anywhere a JWT is accepted.

curl -H "Authorization: Bearer pa_live_your_key_here" \
  http://localhost:8000/api/v1/auth/me

Submit a transcript for the full pipeline. Returns immediately with a job_id when async_mode=true.

curl -X POST http://localhost:8000/api/v1/process \
  -H "Authorization: Bearer $TOKEN" \
  -F "file=@transcript.json" \
  -F "fact_check=true" \
  -F "async_mode=true"

List jobs

curl "http://localhost:8000/api/v1/jobs?status=completed" -H "Authorization: Bearer $TOKEN"

Live SSE stream

import sseclient, requests, json
resp   = requests.get(f"{BASE}/api/v1/jobs/{job_id}/stream", headers=HEADS, stream=True)
client = sseclient.SSEClient(resp)
for event in client.events():
    data = json.loads(event.data)
    if event.event == "pipeline_complete":
        print(data["title"]); break
    print(f"[{event.event}] {data.get('progress',0)}%")

List episodes

curl http://localhost:8000/api/v1/episodes -H "Authorization: Bearer $TOKEN"

Get full report

curl http://localhost:8000/api/v1/episodes/ep001 -H "Authorization: Bearer $TOKEN"

Delete episode

curl -X DELETE http://localhost:8000/api/v1/episodes/ep001 -H "Authorization: Bearer $TOKEN"

Vector search across all indexed episode transcripts. Optionally scope to one episode.

curl -X POST http://localhost:8000/api/v1/search \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"query":"FDA regulation AI","top_k":5}'

RAG-enhanced streaming chat. Returns an OpenAI-compatible SSE stream.

resp = requests.post(f"{BASE}/api/v1/chat",
    json={"messages":[{"role":"user","content":"What FDA claims were verified?"}],
          "episode_id":"ep001"},
    headers=HEADS, stream=True)
for line in resp.iter_lines():
    if line.startswith(b"data: ") and line != b"data: [DONE]":
        import json
        chunk = json.loads(line[6:])["choices"][0]["delta"].get("content","")
        print(chunk, end="", flush=True)

Upload document

curl -X POST http://localhost:8000/api/v1/knowledge -H "Authorization: Bearer $TOKEN" -F "file=@paper.pdf"

Semantic search

curl "http://localhost:8000/api/v1/knowledge/search?q=FDA+AI+approval" -H "Authorization: Bearer $TOKEN"

HTTP status codes

Rate limits

Rate limiting is applied per IP via slowapi. LLM calls are governed by your provider's limits (Groq free tier: ~30 req/min). The pipeline uses exponential backoff with 2 retries per node. Max file sizes: 20 MB per document, 50 MB per transcript.