feat: Rule2 TKG relationship chunks + Phase0-1 Qdrant integration

Phase 0: TKG builder populate face_detections from face.json
- Fix face.json parser for pose_angle format
- Call store_traced_faces.py to set trace_id
- Skip if trace_id already populated

Phase 1: Qdrant face embeddings integration
- Add FaceEmbeddingDb module (src/core/db/face_embedding_db.rs)
- Create dev_face_embeddings collection (dim=512)
- Store 1122 face embeddings with pose metadata
- API: init_collection, batch_upsert, search_similar

Rule2: TKG edges → relationship chunks
- Design: RULE2_TKG_RELATIONSHIP_V1.0.md
- Implementation: rule2_ingest.rs
- ChunkType::Relationship added
- Edge types: SPEAKS_AS, MUTUAL_GAZE, CO_OCCURS_WITH, HAS_APPEARANCE, WEARS
- Auto-trigger on TKG rebuild

API:
- POST /api/v1/file/:file_uuid/rule2 (vectorization)
- POST /api/v1/file/:file_uuid/tkg/rebuild (auto Rule2)

Test: 75 relationship chunks created + vectorized
This commit is contained in:
Accusys
2026-06-21 00:22:41 +08:00
parent 17e4e15860
commit 3ad6f8740a
10 changed files with 3811 additions and 30 deletions

View File

@@ -38,10 +38,17 @@ pub fn trace_agent_routes() -> Router<crate::api::types::AppState> {
get(get_cooccurrence),
)
.route("/api/v1/file/:file_uuid/tkg/rebuild", post(rebuild_tkg))
.route("/api/v1/file/:file_uuid/rule2", post(ingest_rule2))
.route(
"/api/v1/file/:file_uuid/representative-frame",
get(get_representative_frame),
)
.route("/api/v1/file/:file_uuid/tkg/nodes", post(query_tkg_nodes))
.route("/api/v1/file/:file_uuid/tkg/edges", post(query_tkg_edges))
.route(
"/api/v1/file/:file_uuid/tkg/node/:node_id",
get(get_tkg_node_detail),
)
}
#[derive(Debug, Deserialize)]
@@ -961,22 +968,52 @@ async fn rebuild_tkg(
State(state): State<crate::api::types::AppState>,
Path(file_uuid): Path<String>,
) -> Json<TkgRebuildResponse> {
use crate::core::chunk::rule2_ingest::ingest_rule2;
use tracing::info;
let result = crate::core::processor::tkg::build_tkg(&state.db, &file_uuid, &OUTPUT_DIR).await;
match result {
Ok(r) => Json(TkgRebuildResponse {
success: true,
file_uuid,
result: Some(serde_json::json!({
"face_trace_nodes": r.face_trace_nodes,
"object_nodes": r.object_nodes,
"speaker_nodes": r.speaker_nodes,
"co_occurrence_edges": r.co_occurrence_edges,
"speaker_face_edges": r.speaker_face_edges,
"face_face_edges": r.face_face_edges,
})),
error: None,
}),
Ok(r) => {
let total_edges = r.speaker_face_edges
+ r.mutual_gaze_edges
+ r.face_face_edges
+ r.co_occurrence_edges
+ r.has_appearance_edges
+ r.wears_edges;
if total_edges > 0 {
info!("[TKG] {} relationship edges found, triggering Rule 2 ingestion...", total_edges);
match ingest_rule2(state.db.pool(), &file_uuid).await {
Ok(count) => info!("[TKG] Rule 2 created {} relationship chunks", count),
Err(e) => info!("[TKG] Rule 2 ingestion failed: {}", e),
}
}
Json(TkgRebuildResponse {
success: true,
file_uuid,
result: Some(serde_json::json!({
"face_trace_nodes": r.face_trace_nodes,
"gaze_trace_nodes": r.gaze_trace_nodes,
"lip_trace_nodes": r.lip_trace_nodes,
"text_trace_nodes": r.text_trace_nodes,
"appearance_trace_nodes": r.appearance_trace_nodes,
"skin_tone_trace_nodes": r.skin_tone_trace_nodes,
"accessory_nodes": r.accessory_nodes,
"object_nodes": r.object_nodes,
"speaker_nodes": r.speaker_nodes,
"co_occurrence_edges": r.co_occurrence_edges,
"speaker_face_edges": r.speaker_face_edges,
"face_face_edges": r.face_face_edges,
"mutual_gaze_edges": r.mutual_gaze_edges,
"lip_sync_edges": r.lip_sync_edges,
"has_appearance_edges": r.has_appearance_edges,
"wears_edges": r.wears_edges,
})),
error: None,
})
}
Err(e) => Json(TkgRebuildResponse {
success: false,
file_uuid,
@@ -1097,3 +1134,463 @@ async fn get_stranger_thumbnail(
get_trace_thumbnail_inner(&state, &file_uuid, trace_id).await
}
// ── TKG Node/Edge Query APIs ─────────────────────────────────────
fn t(name: &str) -> String {
let schema = std::env::var("DATABASE_SCHEMA").unwrap_or_else(|_| "dev".to_string());
if schema == "public" {
name.to_string()
} else {
format!("{}.{}", schema, name)
}
}
#[derive(Debug, Deserialize)]
struct TkgNodesRequest {
node_type: Option<String>,
page: Option<i64>,
page_size: Option<i64>,
}
#[derive(Debug, Serialize)]
struct TkgNodeInfo {
id: i64,
node_type: String,
external_id: String,
label: String,
properties: serde_json::Value,
}
#[derive(Debug, Serialize)]
struct TkgNodesResponse {
success: bool,
file_uuid: String,
total: i64,
page: i64,
page_size: i64,
nodes: Vec<TkgNodeInfo>,
}
async fn query_tkg_nodes(
State(state): State<crate::api::types::AppState>,
Path(file_uuid): Path<String>,
Json(req): Json<TkgNodesRequest>,
) -> Result<Json<TkgNodesResponse>, (StatusCode, Json<serde_json::Value>)> {
let nodes_table = t("tkg_nodes");
let page = req.page.unwrap_or(1).max(1);
let page_size = req.page_size.unwrap_or(100).max(1).min(500);
let offset = (page - 1) * page_size;
let (where_clause, count_args, query_args) = if let Some(ref node_type) = req.node_type {
(
"WHERE file_uuid = $1 AND node_type = $2".to_string(),
vec![serde_json::json!([&file_uuid, node_type])],
vec![serde_json::json!([
&file_uuid, node_type, page_size, offset
])],
)
} else {
(
"WHERE file_uuid = $1".to_string(),
vec![serde_json::json!([&file_uuid])],
vec![serde_json::json!([&file_uuid, page_size, offset])],
)
};
let total: i64 = if let Some(ref node_type) = req.node_type {
sqlx::query_scalar(&format!(
"SELECT COUNT(*) FROM {} {}",
nodes_table, where_clause
))
.bind(&file_uuid)
.bind(node_type)
.fetch_one(state.db.pool())
.await
.map_err(|e| {
(
StatusCode::INTERNAL_SERVER_ERROR,
Json(serde_json::json!({"error": e.to_string()})),
)
})?
} else {
sqlx::query_scalar(&format!(
"SELECT COUNT(*) FROM {} {}",
nodes_table, where_clause
))
.bind(&file_uuid)
.fetch_one(state.db.pool())
.await
.map_err(|e| {
(
StatusCode::INTERNAL_SERVER_ERROR,
Json(serde_json::json!({"error": e.to_string()})),
)
})?
};
let query = format!(
"SELECT id, node_type, external_id, label, properties FROM {} {} ORDER BY id LIMIT ${} OFFSET ${}",
nodes_table, where_clause,
if req.node_type.is_some() { 3 } else { 2 },
if req.node_type.is_some() { 4 } else { 3 }
);
let rows: Vec<(i64, String, String, String, serde_json::Value)> =
if let Some(ref node_type) = req.node_type {
sqlx::query_as(&query)
.bind(&file_uuid)
.bind(node_type)
.bind(page_size)
.bind(offset)
.fetch_all(state.db.pool())
.await
.map_err(|e| {
(
StatusCode::INTERNAL_SERVER_ERROR,
Json(serde_json::json!({"error": e.to_string()})),
)
})?
} else {
sqlx::query_as(&query)
.bind(&file_uuid)
.bind(page_size)
.bind(offset)
.fetch_all(state.db.pool())
.await
.map_err(|e| {
(
StatusCode::INTERNAL_SERVER_ERROR,
Json(serde_json::json!({"error": e.to_string()})),
)
})?
};
let nodes = rows
.into_iter()
.map(
|(id, node_type, external_id, label, properties)| TkgNodeInfo {
id,
node_type,
external_id,
label,
properties,
},
)
.collect();
Ok(Json(TkgNodesResponse {
success: true,
file_uuid,
total,
page,
page_size,
nodes,
}))
}
#[derive(Debug, Deserialize)]
struct TkgEdgesRequest {
edge_type: Option<String>,
source_type: Option<String>,
target_type: Option<String>,
page: Option<i64>,
page_size: Option<i64>,
}
#[derive(Debug, Serialize)]
struct TkgEdgeInfo {
id: i64,
edge_type: String,
source_node_id: i64,
target_node_id: i64,
properties: serde_json::Value,
}
#[derive(Debug, Serialize)]
struct TkgEdgesResponse {
success: bool,
file_uuid: String,
total: i64,
page: i64,
page_size: i64,
edges: Vec<TkgEdgeInfo>,
}
async fn query_tkg_edges(
State(state): State<crate::api::types::AppState>,
Path(file_uuid): Path<String>,
Json(req): Json<TkgEdgesRequest>,
) -> Result<Json<TkgEdgesResponse>, (StatusCode, Json<serde_json::Value>)> {
let edges_table = t("tkg_edges");
let nodes_table = t("tkg_nodes");
let page = req.page.unwrap_or(1).max(1);
let page_size = req.page_size.unwrap_or(100).max(1).min(500);
let offset = (page - 1) * page_size;
let mut conditions = vec!["e.file_uuid = $1".to_string()];
let mut param_idx = 2i32;
let mut joins = String::new();
if let Some(ref edge_type) = req.edge_type {
conditions.push(format!("e.edge_type = ${}", param_idx));
param_idx += 1;
}
if req.source_type.is_some() || req.target_type.is_some() {
joins = format!(
" JOIN {} sn ON e.source_node_id = sn.id JOIN {} tn ON e.target_node_id = tn.id",
nodes_table, nodes_table
);
}
if let Some(ref source_type) = req.source_type {
conditions.push(format!("sn.node_type = ${}", param_idx));
param_idx += 1;
}
if let Some(ref target_type) = req.target_type {
conditions.push(format!("tn.node_type = ${}", param_idx));
param_idx += 1;
}
let where_clause = conditions.join(" AND ");
let count_query = format!(
"SELECT COUNT(*) FROM {} e {} WHERE {}",
edges_table, joins, where_clause
);
let total: i64 = {
let mut q = sqlx::query_scalar::<_, i64>(&count_query).bind(&file_uuid);
if let Some(ref edge_type) = req.edge_type {
q = q.bind(edge_type);
}
if let Some(ref source_type) = req.source_type {
q = q.bind(source_type);
}
if let Some(ref target_type) = req.target_type {
q = q.bind(target_type);
}
q.fetch_one(state.db.pool()).await.map_err(|e| {
(
StatusCode::INTERNAL_SERVER_ERROR,
Json(serde_json::json!({"error": e.to_string()})),
)
})?
};
let query = format!(
"SELECT e.id, e.edge_type, e.source_node_id, e.target_node_id, e.properties FROM {} e {} WHERE {} ORDER BY e.id LIMIT ${} OFFSET ${}",
edges_table, joins, where_clause, param_idx, param_idx + 1
);
let rows: Vec<(i64, String, i64, i64, serde_json::Value)> = {
let mut q = sqlx::query_as(&query).bind(&file_uuid);
if let Some(ref edge_type) = req.edge_type {
q = q.bind(edge_type);
}
if let Some(ref source_type) = req.source_type {
q = q.bind(source_type);
}
if let Some(ref target_type) = req.target_type {
q = q.bind(target_type);
}
q.bind(page_size)
.bind(offset)
.fetch_all(state.db.pool())
.await
.map_err(|e| {
(
StatusCode::INTERNAL_SERVER_ERROR,
Json(serde_json::json!({"error": e.to_string()})),
)
})?
};
let edges = rows
.into_iter()
.map(
|(id, edge_type, source_node_id, target_node_id, properties)| TkgEdgeInfo {
id,
edge_type,
source_node_id,
target_node_id,
properties,
},
)
.collect();
Ok(Json(TkgEdgesResponse {
success: true,
file_uuid,
total,
page,
page_size,
edges,
}))
}
#[derive(Debug, Serialize)]
struct TkgNodeWithEdges {
node: TkgNodeInfo,
incoming_edges: Vec<TkgEdgeInfo>,
outgoing_edges: Vec<TkgEdgeInfo>,
}
#[derive(Debug, Serialize)]
struct TkgNodeDetailResponse {
success: bool,
file_uuid: String,
node: Option<TkgNodeWithEdges>,
error: Option<String>,
}
async fn get_tkg_node_detail(
State(state): State<crate::api::types::AppState>,
Path((file_uuid, node_id)): Path<(String, i64)>,
) -> Json<TkgNodeDetailResponse> {
let nodes_table = t("tkg_nodes");
let edges_table = t("tkg_edges");
let node: Option<(i64, String, String, String, serde_json::Value)> = sqlx::query_as(
&format!("SELECT id, node_type, external_id, label, properties FROM {} WHERE file_uuid = $1 AND id = $2", nodes_table)
)
.bind(&file_uuid).bind(node_id)
.fetch_optional(state.db.pool()).await.ok().flatten();
match node {
Some((id, node_type, external_id, label, properties)) => {
let incoming: Vec<TkgEdgeInfo> = sqlx::query_as(
&format!("SELECT id, edge_type, source_node_id, target_node_id, properties FROM {} WHERE file_uuid = $1 AND target_node_id = $2", edges_table)
)
.bind(&file_uuid).bind(node_id)
.fetch_all(state.db.pool()).await.unwrap_or_default()
.into_iter().map(|(id, edge_type, source_node_id, target_node_id, properties)| {
TkgEdgeInfo { id, edge_type, source_node_id, target_node_id, properties }
}).collect();
let outgoing: Vec<TkgEdgeInfo> = sqlx::query_as(
&format!("SELECT id, edge_type, source_node_id, target_node_id, properties FROM {} WHERE file_uuid = $1 AND source_node_id = $2", edges_table)
)
.bind(&file_uuid).bind(node_id)
.fetch_all(state.db.pool()).await.unwrap_or_default()
.into_iter().map(|(id, edge_type, source_node_id, target_node_id, properties)| {
TkgEdgeInfo { id, edge_type, source_node_id, target_node_id, properties }
}).collect();
Json(TkgNodeDetailResponse {
success: true,
file_uuid,
node: Some(TkgNodeWithEdges {
node: TkgNodeInfo {
id,
node_type,
external_id,
label,
properties,
},
incoming_edges: incoming,
outgoing_edges: outgoing,
}),
error: None,
})
}
None => Json(TkgNodeDetailResponse {
success: false,
file_uuid,
node: None,
error: Some("Node not found".to_string()),
}),
}
}
// ── Rule 2 Ingest ───────────────────────────────────────────────────
#[derive(Serialize)]
struct IngestRule2Response {
success: bool,
file_uuid: String,
rule2_chunks: i64,
vectorized_chunks: Option<i64>,
error: Option<String>,
}
async fn ingest_rule2(
State(state): State<crate::api::types::AppState>,
Path(file_uuid): Path<String>,
) -> Result<Json<IngestRule2Response>, (StatusCode, Json<serde_json::Value>)> {
use crate::core::chunk::rule2_ingest::ingest_rule2;
use crate::core::embedding::Embedder;
use crate::core::db::schema;
use crate::core::db::qdrant_db::{QdrantDb, VectorPayload};
use tracing::info;
let result = ingest_rule2(state.db.pool(), &file_uuid).await;
match result {
Ok(rule2_chunks) => {
info!(
"[Rule2API] {} relationship chunks created for {}",
rule2_chunks, file_uuid
);
// Auto-vectorize relationship chunks
let embedder = Embedder::new("embeddinggemma-300m".to_string());
let qdrant = QdrantDb::new();
let pool = state.db.pool();
let chunk_table = schema::table_name("chunk");
let rows: Vec<(String, String, i64, i64, f64, f64)> = sqlx::query_as(&format!(
"SELECT chunk_id, text_content, start_frame, end_frame, start_time, end_time \
FROM {} WHERE file_uuid = $1 AND chunk_type = 'relationship' \
AND embedding IS NULL AND (text_content IS NOT NULL AND text_content != '')",
chunk_table
))
.bind(&file_uuid)
.fetch_all(pool)
.await
.map_err(|e| {
(
StatusCode::INTERNAL_SERVER_ERROR,
Json(serde_json::json!({"error": e.to_string()})),
)
})?;
let mut vectorized = 0i64;
for (chunk_id, text, start_frame, end_frame, start_time, end_time) in &rows {
if text.is_empty() {
continue;
}
if let Ok(vector) = embedder.embed_document(&text).await {
if state.db.store_vector(&chunk_id, &vector, &file_uuid).await.is_ok() {
let payload = VectorPayload {
file_uuid: file_uuid.clone(),
chunk_id: chunk_id.clone(),
chunk_type: "relationship".to_string(),
start_frame: *start_frame,
end_frame: *end_frame,
start_time: *start_time,
end_time: *end_time,
text: Some(text.clone()),
};
if qdrant.upsert_vector(&chunk_id, &vector, payload).await.is_ok() {
vectorized += 1;
}
}
}
}
Ok(Json(IngestRule2Response {
success: true,
file_uuid,
rule2_chunks: rule2_chunks as i64,
vectorized_chunks: Some(vectorized),
error: None,
}))
}
Err(e) => Ok(Json(IngestRule2Response {
success: false,
file_uuid,
rule2_chunks: 0,
vectorized_chunks: None,
error: Some(e.to_string()),
})),
}
}