Use strict OpenRouter schema for analysis

.gitea/workflows/build-and-deploy.yml

@@ -0,0 +1,60 @@
+name: Build and deploy Worker
+
+on:
+  push:
+    branches:
+      - main
+
+jobs:
+  build:
+    runs-on: builder
+    env:
+      SERVICE_NAME: worker
+      IMAGE_SHA: gitea.dsrptlab.com/mapflow/worker:${{ github.sha }}
+      IMAGE_LATEST: gitea.dsrptlab.com/mapflow/worker:latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Configure Gitea registry auth
+        run: |
+          set -euo pipefail
+          mkdir -p ~/.docker
+          auth="$(printf '%s:%s' "${{ secrets.REGISTRY_USERNAME }}" "${{ secrets.REGISTRY_TOKEN }}" | base64 | tr -d '\n')"
+          printf '{"auths":{"gitea.dsrptlab.com":{"auth":"%s"}}}\n' "$auth" > ~/.docker/config.json
+
+      - name: Build and push image
+        run: |
+          set -euo pipefail
+          for attempt in 1 2 3; do
+            if docker buildx build --push --provenance=false --tag "$IMAGE_SHA" --tag "$IMAGE_LATEST" .; then
+              exit 0
+            fi
+            sleep "$((attempt * 10))"
+          done
+          exit 1
+
+      - name: Skip stale deployment
+        run: |
+          set -euo pipefail
+          latest_sha="$(git ls-remote origin refs/heads/main | awk '{print $1}')"
+          if [ "$latest_sha" = "${GITHUB_SHA}" ]; then
+            touch .deploy-current
+          else
+            echo "A newer main commit exists: $latest_sha. Skipping deploy for ${GITHUB_SHA}."
+          fi
+
+      - name: Trigger Dokploy deploy webhook
+        run: |
+          set -euo pipefail
+          [ -f .deploy-current ] || exit 0
+          payload=$(cat <<JSON
+          {"ref":"refs/heads/main","after":"$GITHUB_SHA","commits":[{"id":"$GITHUB_SHA","message":"$SERVICE_NAME #${GITHUB_RUN_NUMBER:-0} ${GITHUB_SHA:0:7}","modified":["Dockerfile"]}]}
+          JSON
+          )
+          response_file="$(mktemp)"
+          status_code="$(curl -sS -o "$response_file" -w "%{http_code}" -X POST "${{ secrets.DOKPLOY_DEPLOY_WEBHOOK }}" \
+            -H "x-gitea-event: push" \
+            -H "Content-Type: application/json" \
+            -d "$payload")"
+          cat "$response_file"
+          [ "$status_code" = "200" ]

Dockerfile

@@ -5,10 +5,9 @@ RUN npm ci
 
 FROM node:22-alpine AS build
 WORKDIR /app
-ENV DATABASE_URL="postgresql://mapflow:mapflow@localhost:5432/mapflow"
 COPY --from=deps /app/node_modules ./node_modules
 COPY . .
-RUN npm run prisma:generate
+RUN DATABASE_URL="postgresql://mapflow:mapflow@localhost:5432/mapflow" npm run prisma:generate
 RUN npm run build
 
 FROM node:22-alpine AS runner
@@ -17,7 +16,7 @@ ENV NODE_ENV=production
 COPY package*.json ./
 RUN npm ci --omit=dev
 COPY --from=build /app/dist ./dist
-COPY --from=build /app/src/generated ./src/generated
+COPY --from=build /app/src/generated ./dist/generated
 COPY --from=build /app/prisma ./prisma
 COPY --from=build /app/prisma.config.ts ./prisma.config.ts
 CMD ["node", "dist/hatchet/worker.js"]

prisma/schema.prisma

@@ -17,25 +17,46 @@ enum VoiceExperienceStatus {
 }
 
 model Place {
-  id            String            @id @default(cuid())
-  googlePlaceId String            @unique
-  name          String
-  latitude      Float
-  longitude     Float
-  experiences   VoiceExperience[]
-  createdAt     DateTime          @default(now())
-  updatedAt     DateTime          @updatedAt
+  id                String            @id @default(cuid())
+  googlePlaceId     String            @unique
+  name              String
+  latitude          Float
+  longitude         Float
+  googlePrimaryType String?
+  googleTypes       String[]          @default([])
+  experiences       VoiceExperience[]
+  createdAt         DateTime          @default(now())
+  updatedAt         DateTime          @updatedAt
+}
+
+model User {
+  id               String            @id @default(cuid())
+  telegramId       String            @unique
+  username         String?
+  firstName        String?
+  lastName         String?
+  photoUrl         String?
+  languageCode     String?
+  isAdmin          Boolean           @default(false)
+  voiceExperiences VoiceExperience[]
+  createdAt        DateTime          @default(now())
+  updatedAt        DateTime          @updatedAt
 }
 
 model VoiceExperience {
-  id              String                @id @default(cuid())
-  placeId          String
-  place            Place                 @relation(fields: [placeId], references: [id])
-  durationSeconds  Int
-  audioObjectKey   String
-  status           VoiceExperienceStatus @default(UPLOADED)
-  transcript       String?
-  analysis         Json?
-  createdAt        DateTime              @default(now())
-  updatedAt        DateTime              @updatedAt
+  id                 String                @id @default(cuid())
+  placeId            String
+  place              Place                 @relation(fields: [placeId], references: [id])
+  userId             String?
+  user               User?                 @relation(fields: [userId], references: [id])
+  durationSeconds    Int
+  audioObjectKey     String
+  audioContentBase64 String?
+  audioMimeType      String?
+  audioAccessToken   String?               @unique
+  status             VoiceExperienceStatus @default(UPLOADED)
+  transcript         String?
+  analysis           Json?
+  createdAt          DateTime              @default(now())
+  updatedAt          DateTime              @updatedAt
 }

src/config.ts

@@ -1,7 +1,18 @@
 import 'dotenv/config';
 
+import { loadVaultEnvironment } from './vault/env.js';
+
+await loadVaultEnvironment();
+
 export const config = {
   databaseUrl: process.env.DATABASE_URL ?? '',
   workerName: process.env.HATCHET_WORKER_NAME ?? 'mapflow-hatchet-worker',
   workerSlots: Number.parseInt(process.env.HATCHET_WORKER_SLOTS ?? '4', 10),
+  publicApiUrl: process.env.PUBLIC_API_URL ?? 'https://api.map.craftee.vn',
+  deepgramApiKey: process.env.DEEPGRAM_API_KEY ?? '',
+  deepgramModel: process.env.DEEPGRAM_MODEL ?? 'nova-3',
+  deepgramLanguage: process.env.DEEPGRAM_LANGUAGE ?? 'ru',
+  openRouterApiKey: process.env.OPENROUTER_API_KEY ?? '',
+  openRouterModel:
+    process.env.OPENROUTER_MODEL ?? 'deepseek/deepseek-v4-flash',
 };

src/generated/prisma/index-browser.js

@@ -126,6 +126,21 @@ exports.Prisma.PlaceScalarFieldEnum = {
   name: 'name',
   latitude: 'latitude',
   longitude: 'longitude',
+  googlePrimaryType: 'googlePrimaryType',
+  googleTypes: 'googleTypes',
+  createdAt: 'createdAt',
+  updatedAt: 'updatedAt'
+};
+
+exports.Prisma.UserScalarFieldEnum = {
+  id: 'id',
+  telegramId: 'telegramId',
+  username: 'username',
+  firstName: 'firstName',
+  lastName: 'lastName',
+  photoUrl: 'photoUrl',
+  languageCode: 'languageCode',
+  isAdmin: 'isAdmin',
   createdAt: 'createdAt',
   updatedAt: 'updatedAt'
 };
@@ -133,8 +148,12 @@ exports.Prisma.PlaceScalarFieldEnum = {
 exports.Prisma.VoiceExperienceScalarFieldEnum = {
   id: 'id',
   placeId: 'placeId',
+  userId: 'userId',
   durationSeconds: 'durationSeconds',
   audioObjectKey: 'audioObjectKey',
+  audioContentBase64: 'audioContentBase64',
+  audioMimeType: 'audioMimeType',
+  audioAccessToken: 'audioAccessToken',
   status: 'status',
   transcript: 'transcript',
   analysis: 'analysis',
@@ -157,16 +176,16 @@ exports.Prisma.QueryMode = {
   insensitive: 'insensitive'
 };
 
+exports.Prisma.NullsOrder = {
+  first: 'first',
+  last: 'last'
+};
+
 exports.Prisma.JsonNullValueFilter = {
   DbNull: Prisma.DbNull,
   JsonNull: Prisma.JsonNull,
   AnyNull: Prisma.AnyNull
 };
-
-exports.Prisma.NullsOrder = {
-  first: 'first',
-  last: 'last'
-};
 exports.VoiceExperienceStatus = exports.$Enums.VoiceExperienceStatus = {
   UPLOADED: 'UPLOADED',
   TRANSCRIBING: 'TRANSCRIBING',
@@ -178,6 +197,7 @@ exports.VoiceExperienceStatus = exports.$Enums.VoiceExperienceStatus = {
 
 exports.Prisma.ModelName = {
   Place: 'Place',
+  User: 'User',
   VoiceExperience: 'VoiceExperience'
 };
 

src/generated/prisma/package.json

@@ -1,5 +1,5 @@
 {
-  "name": "prisma-client-5316d8e66293a4954be6e26169692366997bf25fe9186e98c87e62eed3dc691e",
+  "name": "prisma-client-404088dc715856cb4ed2e7f2d4b8eb35c87427ce32c0e67bc04d6ff67961fc2b",
   "main": "index.js",
   "types": "index.d.ts",
   "browser": "default.js",

src/generated/prisma/schema.prisma

@@ -17,25 +17,46 @@ enum VoiceExperienceStatus {
 }
 
 model Place {
-  id            String            @id @default(cuid())
-  googlePlaceId String            @unique
-  name          String
-  latitude      Float
-  longitude     Float
-  experiences   VoiceExperience[]
-  createdAt     DateTime          @default(now())
-  updatedAt     DateTime          @updatedAt
+  id                String            @id @default(cuid())
+  googlePlaceId     String            @unique
+  name              String
+  latitude          Float
+  longitude         Float
+  googlePrimaryType String?
+  googleTypes       String[]          @default([])
+  experiences       VoiceExperience[]
+  createdAt         DateTime          @default(now())
+  updatedAt         DateTime          @updatedAt
+}
+
+model User {
+  id               String            @id @default(cuid())
+  telegramId       String            @unique
+  username         String?
+  firstName        String?
+  lastName         String?
+  photoUrl         String?
+  languageCode     String?
+  isAdmin          Boolean           @default(false)
+  voiceExperiences VoiceExperience[]
+  createdAt        DateTime          @default(now())
+  updatedAt        DateTime          @updatedAt
 }
 
 model VoiceExperience {
-  id              String                @id @default(cuid())
-  placeId         String
-  place           Place                 @relation(fields: [placeId], references: [id])
-  durationSeconds Int
-  audioObjectKey  String
-  status          VoiceExperienceStatus @default(UPLOADED)
-  transcript      String?
-  analysis        Json?
-  createdAt       DateTime              @default(now())
-  updatedAt       DateTime              @updatedAt
+  id                 String                @id @default(cuid())
+  placeId            String
+  place              Place                 @relation(fields: [placeId], references: [id])
+  userId             String?
+  user               User?                 @relation(fields: [userId], references: [id])
+  durationSeconds    Int
+  audioObjectKey     String
+  audioContentBase64 String?
+  audioMimeType      String?
+  audioAccessToken   String?               @unique
+  status             VoiceExperienceStatus @default(UPLOADED)
+  transcript         String?
+  analysis           Json?
+  createdAt          DateTime              @default(now())
+  updatedAt          DateTime              @updatedAt
 }

src/hatchet/worker.ts

@@ -1,8 +1,6 @@
 import 'dotenv/config';
 
 import { config } from '../config.js';
-import { hatchet } from './hatchet-client.js';
-import { processVoiceExperienceWorkflow } from './workflows/process-voice-experience.js';
 
 function resolveWorkerSlots(): number {
   if (!Number.isFinite(config.workerSlots) || config.workerSlots <= 0) return 4;
@@ -14,6 +12,10 @@ async function main() {
     throw new Error('HATCHET_CLIENT_TOKEN is required for hatchet worker');
   }
 
+  const { hatchet } = await import('./hatchet-client.js');
+  const { processVoiceExperienceWorkflow } = await import(
+    './workflows/process-voice-experience.js'
+  );
   const worker = await hatchet.worker(config.workerName, {
     workflows: [processVoiceExperienceWorkflow],
     slots: resolveWorkerSlots(),

src/jobs/analyze-voice-experience.ts

@@ -1,6 +1,31 @@
-import { buildPlaceAnalysis } from '../ontology/place-ontology.js';
+import { config } from '../config.js';
+import { placeOntology } from '../ontology/place-ontology.js';
 import { prisma } from '../prisma.js';
 
+type OpenRouterResponse = {
+  choices?: Array<{
+    finish_reason?: string;
+    message?: {
+      content?: string;
+    };
+  }>;
+  error?: {
+    message?: string;
+  };
+};
+
+type PlaceAnalysis = {
+  placeName: string;
+  tags: string[];
+  signals: Array<{
+    axis: string;
+    leaf: string;
+    evidence: string;
+    confidence: number;
+  }>;
+  summary: string;
+};
+
 export async function analyzeVoiceExperience(experienceId: string) {
   const experience = await prisma.voiceExperience.findUnique({
     where: { id: experienceId },
@@ -20,7 +45,7 @@ export async function analyzeVoiceExperience(experienceId: string) {
     data: { status: 'ANALYZING' },
   });
 
-  const analysis = buildPlaceAnalysis({
+  const analysis = await buildPlaceAnalysis({
     placeName: experience.place.name,
     transcript: experience.transcript,
   });
@@ -35,3 +60,136 @@ export async function analyzeVoiceExperience(experienceId: string) {
 
   return { tags: analysis.tags };
 }
+
+async function buildPlaceAnalysis(input: {
+  placeName: string;
+  transcript: string;
+}): Promise<PlaceAnalysis> {
+  if (!config.openRouterApiKey) {
+    throw new Error('OPENROUTER_API_KEY is required for voice analysis.');
+  }
+
+  const ontology = placeOntology.map((axis) => ({
+    axis: axis.id,
+    options: axis.leaves.map((leaf) => ({
+      tag: `${axis.id}:${leaf.id}`,
+      keywords: leaf.keywords,
+    })),
+  }));
+  const response = await fetch('https://openrouter.ai/api/v1/chat/completions', {
+    method: 'POST',
+    headers: {
+      authorization: `Bearer ${config.openRouterApiKey}`,
+      'content-type': 'application/json',
+      'http-referer': 'https://map.craftee.vn',
+      'x-title': 'MapFlow',
+    },
+    body: JSON.stringify({
+      model: config.openRouterModel,
+      provider: { require_parameters: true },
+      temperature: 0.1,
+      max_tokens: 2000,
+      response_format: {
+        type: 'json_schema',
+        json_schema: {
+          name: 'place_analysis',
+          strict: true,
+          schema: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['placeName', 'tags', 'signals', 'summary'],
+            properties: {
+              placeName: { type: 'string' },
+              tags: { type: 'array', items: { type: 'string' } },
+              signals: {
+                type: 'array',
+                items: {
+                  type: 'object',
+                  additionalProperties: false,
+                  required: ['axis', 'leaf', 'evidence', 'confidence'],
+                  properties: {
+                    axis: { type: 'string' },
+                    leaf: { type: 'string' },
+                    evidence: { type: 'string' },
+                    confidence: { type: 'number', minimum: 0, maximum: 1 },
+                  },
+                },
+              },
+              summary: { type: 'string' },
+            },
+          },
+        },
+      },
+      messages: [
+        {
+          role: 'system',
+          content:
+            'You classify voice reviews about places. Return one valid JSON object in message.content. Do not return markdown. Use only allowed ontology tags.',
+        },
+        {
+          role: 'user',
+          content: JSON.stringify({
+            task:
+              'Analyze the transcript and select place-experience tags that are explicitly supported by the text. Do not invent facts.',
+            outputShape: {
+              placeName: 'string',
+              tags: ['axis:leaf'],
+              signals: [
+                {
+                  axis: 'string',
+                  leaf: 'string',
+                  evidence: 'short quote or paraphrase from transcript',
+                  confidence: 'number from 0 to 1',
+                },
+              ],
+              summary: 'one short sentence',
+            },
+            placeName: input.placeName,
+            ontology,
+            transcript: input.transcript,
+          }),
+        },
+      ],
+    }),
+  });
+
+  if (!response.ok) {
+    const errorBody = await response.text();
+    throw new Error(
+      `OpenRouter analysis failed with ${response.status}: ${errorBody.slice(0, 500)}`,
+    );
+  }
+
+  const payload = (await response.json()) as OpenRouterResponse;
+  const choice = payload.choices?.[0];
+  const content = choice?.message?.content?.trim();
+  if (!content) {
+    throw new Error(
+      `OpenRouter returned an empty analysis for ${config.openRouterModel}; finish_reason=${choice?.finish_reason ?? 'missing'}.`,
+    );
+  }
+
+  const analysis = JSON.parse(content) as PlaceAnalysis;
+  assertValidAnalysis(analysis);
+  return analysis;
+}
+
+function assertValidAnalysis(analysis: PlaceAnalysis) {
+  const allowedTags = new Set(
+    placeOntology.flatMap((axis) =>
+      axis.leaves.map((leaf) => `${axis.id}:${leaf.id}`),
+    ),
+  );
+
+  if (!Array.isArray(analysis.tags)) {
+    throw new Error('OpenRouter analysis tags must be an array.');
+  }
+  for (const tag of analysis.tags) {
+    if (!allowedTags.has(tag)) {
+      throw new Error(`OpenRouter returned unsupported tag: ${tag}.`);
+    }
+  }
+  if (!Array.isArray(analysis.signals)) {
+    throw new Error('OpenRouter analysis signals must be an array.');
+  }
+}

src/jobs/transcribe-voice-experience.ts

@@ -1,5 +1,19 @@
+import { config } from '../config.js';
 import { prisma } from '../prisma.js';
 
+type DeepgramResponse = {
+  results?: {
+    channels?: Array<{
+      alternatives?: Array<{
+        transcript?: string;
+        paragraphs?: {
+          transcript?: string;
+        };
+      }>;
+    }>;
+  };
+};
+
 export async function transcribeVoiceExperience(experienceId: string) {
   const experience = await prisma.voiceExperience.findUnique({
     where: { id: experienceId },
@@ -14,11 +28,13 @@ export async function transcribeVoiceExperience(experienceId: string) {
     data: { status: 'TRANSCRIBING' },
   });
 
-  const transcript = await transcribeAudioObject(experience.audioObjectKey);
+  const transcript = await transcribeAudioObject(experience.id);
 
   await prisma.voiceExperience.update({
     where: { id: experienceId },
     data: {
+      audioContentBase64: null,
+      audioAccessToken: null,
       transcript,
       status: 'TRANSCRIBED',
     },
@@ -27,7 +43,68 @@ export async function transcribeVoiceExperience(experienceId: string) {
   return { transcript };
 }
 
-async function transcribeAudioObject(audioObjectKey: string): Promise<string> {
-  // TODO: replace this adapter with the production speech-to-text provider.
-  return `Transcription placeholder for ${audioObjectKey}`;
+async function transcribeAudioObject(experienceId: string): Promise<string> {
+  if (!config.deepgramApiKey) {
+    throw new Error('DEEPGRAM_API_KEY is required for transcription.');
+  }
+
+  const experience = await prisma.voiceExperience.findUnique({
+    where: { id: experienceId },
+    select: {
+      id: true,
+      audioObjectKey: true,
+      audioAccessToken: true,
+      audioContentBase64: true,
+      audioMimeType: true,
+    },
+  });
+
+  if (!experience) {
+    throw new Error(`Voice experience ${experienceId} was not found.`);
+  }
+  if (!experience.audioAccessToken) {
+    throw new Error(
+      `Voice audio ${experience.audioObjectKey} has no access token.`,
+    );
+  }
+  if (!experience.audioContentBase64 || !experience.audioMimeType) {
+    throw new Error(
+      `Voice audio ${experience.audioObjectKey} has no stored content.`,
+    );
+  }
+
+  const params = new URLSearchParams({
+    model: config.deepgramModel,
+    language: config.deepgramLanguage,
+    smart_format: 'true',
+    punctuate: 'true',
+  });
+  const audioUrl = `${config.publicApiUrl.replace(/\/$/, '')}/audio/voice-experiences/${experience.id}?token=${encodeURIComponent(experience.audioAccessToken)}`;
+  const response = await fetch(
+    `https://api.deepgram.com/v1/listen?${params.toString()}`,
+    {
+      method: 'POST',
+      headers: {
+        authorization: `Token ${config.deepgramApiKey}`,
+        'content-type': 'application/json',
+      },
+      body: JSON.stringify({ url: audioUrl }),
+    },
+  );
+
+  if (!response.ok) {
+    throw new Error(`Deepgram transcription failed with ${response.status}.`);
+  }
+
+  const payload = (await response.json()) as DeepgramResponse;
+  const alternative = payload.results?.channels?.[0]?.alternatives?.[0];
+  const transcript =
+    alternative?.paragraphs?.transcript?.trim() ??
+    alternative?.transcript?.trim() ??
+    '';
+  if (!transcript) {
+    throw new Error('Deepgram returned an empty transcript.');
+  }
+
+  return transcript;
 }

src/ontology/place-ontology.ts

@@ -31,7 +31,7 @@ export const placeOntology: OntologyAxis[] = [
     ],
   },
   {
-    id: 'intent',
+    id: 'function',
     leaves: [
       { id: 'reset', keywords: ['rest', 'breathe', 'reset', 'recover'] },
       { id: 'impress', keywords: ['impress', 'beautiful', 'special', 'wow'] },

src/vault/env.ts

@@ -0,0 +1,65 @@
+type VaultConfig = {
+  address: string;
+  token: string;
+  mount: string;
+  sharedPath: string;
+  projectPath: string;
+};
+
+function requireEnv(name: string) {
+  const value = process.env[name];
+  if (!value) {
+    throw new Error(`${name} is required when VAULT_ENABLED=true.`);
+  }
+  return value;
+}
+
+function vaultConfig(): VaultConfig {
+  return {
+    address: requireEnv('VAULT_ADDR').replace(/\/$/, ''),
+    token: requireEnv('VAULT_TOKEN'),
+    mount: requireEnv('VAULT_KV_MOUNT'),
+    sharedPath: requireEnv('VAULT_SHARED_PATH'),
+    projectPath: requireEnv('VAULT_PROJECT_PATH'),
+  };
+}
+
+async function readVaultPath(config: VaultConfig, path: string) {
+  const response = await fetch(
+    `${config.address}/v1/${config.mount}/data/${path}`,
+    { headers: { 'X-Vault-Token': config.token } },
+  );
+
+  if (!response.ok) {
+    throw new Error(`Vault read failed for ${path}: ${response.status}.`);
+  }
+
+  const payload = (await response.json()) as {
+    data?: { data?: Record<string, unknown> };
+  };
+  const data = payload.data?.data;
+  if (!data) {
+    throw new Error(`Vault path ${path} has no KV v2 data.`);
+  }
+
+  return data;
+}
+
+function applyEnvironment(values: Record<string, unknown>) {
+  for (const [key, value] of Object.entries(values)) {
+    if (typeof value !== 'string') {
+      throw new Error(`Vault value ${key} must be a string.`);
+    }
+    process.env[key] = value;
+  }
+}
+
+export async function loadVaultEnvironment() {
+  if (process.env.VAULT_ENABLED !== 'true') {
+    return;
+  }
+
+  const config = vaultConfig();
+  applyEnvironment(await readVaultPath(config, config.sharedPath));
+  applyEnvironment(await readVaultPath(config, config.projectPath));
+}