# Project export: The First Match This document was generated by HackStack to give an AI agent context about a hackathon project. Sections are labeled with their provenance; content marked as truncated was cut to keep this document small. ## Project metadata - Hackathon: TreeHacks 2026 - Tagline: The only foster matching system that allows children to speak for themselves. - Devpost: https://devpost.com/software/the-first-match - GitHub: https://github.com/ramlukn/treehacks - Team: 2 GitHub contributor(s) — ramlukn (3 commits), ramlukn (2 commits) ## Devpost submission (written by the team) ### Inspiration Every year, over 400,000 children sit in the U.S. foster care system — and nearly 50% of placements disrupt. Caseworkers juggling 20+ cases rely on gut instinct and spreadsheets to make life-altering matches. Nikhil has extensive experience in developmental therapy through his startup SportsMind, working intensively with high-performing athletes on psychological resilience. Julio has worked in Ed and now EdTech, specifically with students from marginalized contexts. In a conversation about one of Julio's previous students, the foster care system came up — how flawed the matching process is. Julio saw firsthand the severity that a familial mismatch could lead to: behavioral regression, broken trust, and a child who becomes harder to place with every failed attempt. They decided to build the solution themselves. What It Does Sunny, the AI Counselor — Children talk to a warm, live video avatar (HeyGen + Claude) instead of filling out clinical forms. Sunny adapts to the child's age, extracts 30+ psychosocial dimensions and 6 safety flags from a natural 5-minute conversation. No clipboards. Family Profiling — Families complete guided onboarding capturing parenting style, emotional capacity, trauma training, cultural practices, and environment details. Caseworker Note Intelligence — Caseworkers paste free-text notes; Claude extracts structured clinical metrics and cross-references them with the AI counselor's findings. 7-Layer Matching — Every child-family pair runs through a 1,142-line matching engine: $$\text{MatchScore} = \sum_{i=1}^{7} w_i \cdot L_i(\text{child}, \text{family}, \text{context})$$ L1: Profile Normalization L2: Hard Constraints L3: 7 Weighted Dimensions (Personality, Emotional, Attachment, Cultural, Communication, Environment, Stability) L4: Cross-Dimensional Synergy L5: Contextual Reasoning (dual-source: AI transcripts + caseworker notes) L6: Bias Audit & Fairness L7: Stability Prediction & Risk Classification Vector pre-filtering uses pgvector with custom 25-dimensional embeddings (cosine similarity) to narrow candidates before full reasoning. How We Built It Frontend: Flutter web, Provider state management, 10 screens AI Counselor: Claude Sonnet 4.5 (7 conversation phases) + HeyGen Streaming Avatar (LiveKit/WebRTC) + speech-to-text/TTS with auto-listen Backend: Supabase — PostgreSQL + pgvector, 6 tables, 3 Edge Functions (Deno) Matching: 7-layer engine running server-side via Edge Function, local Dart fallback Embeddings: 25-dim vectors encoding energy, attachment, trust, culture, environment, and 20 other factors Challenges We Ran Into Ordering problem: Transcripts saved before the child existed in DB, violating FK constraints. Had to restructure the entire UUID/save flow. Client vs. server matching: Built the engine client-side first, then migrated to Edge Functions for database access while keeping local fallback. Avatar + speech coordination: HeyGen video, STT, and TTS had to not step on each other — solved with state guards and auto-listen callbacks. Trauma-informed prompting: 300+ line system prompt to make Claude genuinely warm with a 5-year-old while extracting clinical-grade metrics. Algorithmic bias: Built Layer 6 specifically to correct for systematic biases against older children, cross-cultural matches, and special needs. Accomplishments We're Proud Of Sunny feels real — kids don't know they're being assessed Every match has a full transparent audit trail — no black boxes Dual-source validation between AI counselor and caseworker notes Custom 25-dimensional embedding space purpose-built for adoption matching 6 real-time safety flags monitored during every conversation What We Learned The foster care matching problem isn't lack of caring — it's lack of tools Children reveal far more in natural conversation than in checkbox forms Bias correction is core infrastructure, not a nice-to-have — it's literally Layer 6 LLMs can be trauma-informed with the right prompting architecture What's Next for The First Match Longitudinal outcome tracking to improve matching with real-world feedback Multi-language support for Sunny Agency partnerships to pilot against expert caseworker decisions Mobile app for tablets in caseworker offices and group homes Sibling group optimization — keeping siblings together ## README (from the GitHub repository) # treehacks A new Flutter project. ## Getting Started This project is a starting point for a Flutter application. A few resources to get you started if this is your first Flutter project: - [Lab: Write your first Flutter app](https://docs.flutter.dev/get-started/codelab) - [Cookbook: Useful Flutter samples](https://docs.flutter.dev/cookbook) For help getting started with Flutter development, view the [online documentation](https://docs.flutter.dev/), which offers tutorials, samples, guidance on mobile development, and a full API reference. ## Detected evidence (automated analysis) Indexed codebase: 78 recognized source files, 525 KB. - C (language) — detected in the code - C++ (language) — detected in the code - Dart (language) — detected in the code - HTML (language) — detected in the code - JavaScript (language) — detected in the code - Kotlin (language) — detected in the code - Swift (language) — detected in the code - TypeScript (language) — detected in the code - Supabase (technology) — claimed on Devpost, not found in the code ## Codebase structure (from repository index) ### Files (120 of 137) ``` .gitignore .metadata analysis_options.yaml android/.gitignore android/app/build.gradle android/app/src/debug/AndroidManifest.xml android/app/src/main/AndroidManifest.xml android/app/src/main/kotlin/com/example/treehacks/MainActivity.kt android/app/src/main/res/drawable-v21/launch_background.xml android/app/src/main/res/drawable/launch_background.xml android/app/src/main/res/values-night/styles.xml android/app/src/main/res/values/styles.xml android/app/src/profile/AndroidManifest.xml android/build.gradle android/gradle.properties android/gradle/wrapper/gradle-wrapper.properties android/settings.gradle DEVPOST.md ios/.gitignore ios/Flutter/AppFrameworkInfo.plist ios/Flutter/Debug.xcconfig ios/Flutter/Release.xcconfig ios/Podfile ios/Runner.xcodeproj/project.pbxproj ios/Runner.xcodeproj/project.xcworkspace/contents.xcworkspacedata ios/Runner.xcodeproj/project.xcworkspace/xcshareddata/IDEWorkspaceChecks.plist ios/Runner.xcodeproj/project.xcworkspace/xcshareddata/WorkspaceSettings.xcsettings ios/Runner.xcodeproj/xcshareddata/xcschemes/Runner.xcscheme ios/Runner.xcworkspace/contents.xcworkspacedata ios/Runner.xcworkspace/xcshareddata/IDEWorkspaceChecks.plist ios/Runner.xcworkspace/xcshareddata/WorkspaceSettings.xcsettings ios/Runner/AppDelegate.swift ios/Runner/Assets.xcassets/AppIcon.appiconset/Contents.json ios/Runner/Assets.xcassets/LaunchImage.imageset/Contents.json ios/Runner/Assets.xcassets/LaunchImage.imageset/README.md ios/Runner/Base.lproj/LaunchScreen.storyboard ios/Runner/Base.lproj/Main.storyboard ios/Runner/Info.plist ios/Runner/Runner-Bridging-Header.h ios/RunnerTests/RunnerTests.swift lib/config/secrets.example.dart lib/main.dart lib/models/caseworker_notes.dart lib/models/child_profile.dart lib/models/conversation_transcript.dart lib/models/family_profile.dart lib/models/match_result.dart lib/screens/caseworker_notes_screen.dart lib/screens/child_onboarding_screen.dart lib/screens/children_screen.dart lib/screens/counselor_chat_screen.dart lib/screens/dashboard_screen.dart lib/screens/families_screen.dart lib/screens/family_onboarding_screen.dart lib/screens/home_screen.dart lib/screens/match_detail_screen.dart lib/screens/matches_screen.dart lib/services/ai_counselor_service.dart lib/services/app_state.dart lib/services/caseworker_notes_service.dart lib/services/did_service.dart lib/services/elevenlabs_service.dart lib/services/embedding_service.dart lib/services/heygen_service.dart lib/services/matching_engine.dart lib/services/mock_data_service.dart lib/services/speech_service.dart lib/services/supabase_service.dart lib/services/virtual_counselor.dart lib/utils/theme.dart lib/widgets/did_avatar_widget.dart lib/widgets/heygen_avatar_widget.dart lib/widgets/profile_avatar.dart lib/widgets/score_indicator.dart lib/widgets/section_card.dart lib/widgets/sunny_avatar.dart linux/.gitignore linux/CMakeLists.txt linux/flutter/CMakeLists.txt linux/flutter/generated_plugin_registrant.cc linux/flutter/generated_plugin_registrant.h linux/flutter/generated_plugins.cmake linux/runner/CMakeLists.txt linux/runner/main.cc linux/runner/my_application.cc linux/runner/my_application.h macos/.gitignore macos/Flutter/Flutter-Debug.xcconfig macos/Flutter/Flutter-Release.xcconfig macos/Flutter/GeneratedPluginRegistrant.swift macos/Podfile macos/Runner.xcodeproj/project.pbxproj macos/Runner.xcodeproj/project.xcworkspace/xcshareddata/IDEWorkspaceChecks.plist macos/Runner.xcodeproj/xcshareddata/xcschemes/Runner.xcscheme macos/Runner.xcworkspace/contents.xcworkspacedata macos/Runner.xcworkspace/xcshareddata/IDEWorkspaceChecks.plist macos/Runner/AppDelegate.swift macos/Runner/Assets.xcassets/AppIcon.appiconset/Contents.json macos/Runner/Base.lproj/MainMenu.xib macos/Runner/Configs/AppInfo.xcconfig macos/Runner/Configs/Debug.xcconfig macos/Runner/Configs/Release.xcconfig macos/Runner/Configs/Warnings.xcconfig macos/Runner/DebugProfile.entitlements macos/Runner/Info.plist macos/Runner/MainFlutterWindow.swift macos/Runner/Release.entitlements macos/RunnerTests/RunnerTests.swift pubspec.lock pubspec.yaml README.md run_dev.sh supabase/config.toml supabase/functions/compute-matches/index.ts supabase/functions/extract-caseworker-metrics/index.ts supabase/functions/generate-embedding/index.ts web/did_webrtc_bridge.js web/heygen_livekit_bridge.js web/index.html web/manifest.json [17 more files omitted for size] ``` ### Dependencies No dependency index available. ### Recent commits (newest first) - Add files via upload - Merge remote-tracking branch 'origin/master' - "devpost" - Delete lib/config/secrets.dart - "v1" ## Key source files (fetched from GitHub, selected and truncated for size) ### DEVPOST.md ```markdown # The First Match ## Inspiration Every year, over 400,000 children sit in the U.S. foster care system — and nearly **50% of placements disrupt**. Caseworkers juggling 20+ cases rely on gut instinct and spreadsheets to make life-altering matches. Niki has extensive experience in developmental therapy through his startup SportsMind, working intensively with high-performing athletes on psychological resilience. Julio has worked in Ed and now EdTech, specifically with students from marginalized contexts. In a conversation about one of Julio's previous students, the foster care system came up — how flawed the matching process is. Julio saw firsthand the severity that a familial mismatch could lead to: behavioral regression, broken trust, and a child who becomes harder to place with every failed attempt. They decided to build the solution themselves. ## What It Does **Sunny, the AI Counselor** — Children talk to a warm, live video avatar (HeyGen + Claude) instead of filling out clinical forms. Sunny adapts to the child's age, extracts 30+ psychosocial dimensions and 6 safety flags from a natural 5-minute conversation. No clipboards. **Family Profiling** — Families complete guided onboarding capturing parenting style, emotional capacity, trauma training, cultural practices, and environment details. **Caseworker Note Intelligence** — Caseworkers paste free-text notes; Claude extracts structured clinical metrics and cross-references them with the AI counselor's findings. **7-Layer Matching** — Every child-family pair runs through a 1,142-line matching engine: $$\text{MatchScore} = \sum_{i=1}^{7} w_i \cdot L_i(\text{child}, \text{family}, \text{context})$$ L1: Profile Normalization | L2: Hard Constraints | L3: 7 Weighted Dimensions (Personality, Emotional, Attachment, Cultural, Communication, Environment, Stability) | L4: Cross-Dimensional Synergy | L5: Contextual Reasoning (dual-source: AI transcripts + caseworker notes) | L6: Bias Audit & Fairness | L7: Stability Prediction & Risk Classification Vector pre-filtering uses **pgvector** with custom 25-dimensional embeddings (cosine similarity) to narrow candidates before full reasoning. ## How We Built It - **Frontend**: Flutter web, Provider state management, 10 screens - **AI Counselor**: Claude Sonnet 4.5 (7 conversation phases) + HeyGen Streaming Avatar (LiveKit/WebRTC) + speech-to-text/TTS with auto-listen - **Backend**: Supabase — PostgreSQL + pgvector, 6 tables, 3 Edge Functions (Deno) - **Matching**: 7-layer engine running server-side via Edge Function, local Dart fallback - **Embeddings**: 25-dim vectors encoding energy, attachment, trust, culture, environment, and 20 other factors ## Challenges We Ran Into - **Ordering problem**: Transcripts saved before the child existed in DB, violating FK constraints. Had to restructure the entire UUID/save flow. - **Client vs. server matching**: Built the engine client-side first, then migrated to Edge Functions for database access while keeping [truncated — 1491 more characters] ``` ### supabase/functions/generate-embedding/index.ts ```typescript import { serve } from "https://deno.land/std@0.177.0/http/server.ts"; import { createClient } from "https://esm.sh/@supabase/supabase-js@2"; const SUPABASE_URL = Deno.env.get("SUPABASE_URL")!; const SUPABASE_SERVICE_KEY = Deno.env.get("SUPABASE_SERVICE_ROLE_KEY")!; // ---------- 25-Dimension Embedding Helpers ---------- function attachmentToNumeric(style: string): number { const map: Record = { secure: 0.9, anxious: 0.5, avoidant: 0.3, disorganized: 0.1, }; return map[style?.toLowerCase()] ?? 0.5; } function approachToNumeric(approach: string): number { const map: Record = { nurturing: 0.8, "structured-nurturing": 0.6, therapeutic: 0.4, }; return map[approach?.toLowerCase()] ?? 0.5; } function commStyleToWarmth(style: string): number { const map: Record = { warm: 0.9, open: 0.8, direct: 0.5, structured: 0.4, verbal: 0.7, "non-verbal": 0.3, mixed: 0.5, }; return map[style?.toLowerCase()] ?? 0.5; } function envToOneHot(env: string): [number, number, number] { switch (env?.toLowerCase()) { case "urban": return [1, 0, 0]; case "suburban": return [0, 1, 0]; case "rural": return [0, 0, 1]; default: return [0.33, 0.34, 0.33]; } } function hasKeyword(items: string[], keywords: string[]): number { if (!items?.length) return 0; const lower = items.map((i) => i.toLowerCase()); return lower.some((i) => keywords.some((k) => i.includes(k))) ? 1 : 0; } // ---------- Vector Generators ---------- function childToVector(c: Record): number[] { const v = new Array(25).fill(0); v[0] = c.energy_level ?? 0.5; v[1] = (c.social_comfort ?? 0.5) * 0.5 + commStyleToWarmth(c.communication_style) * 0.5; v[2] = c.structure_need ?? 0.5; v[3] = (c.emotional_awareness ?? 0.5) * 0.4 + (1 - (c.trauma_severity ?? 0)) * 0.3 + (c.trust_level ?? 0.5) * 0.3; v[4] = attachmentToNumeric(c.attachment_style); v[5] = 1 - (c.trauma_severity ?? 0); v[6] = c.trust_level ?? 0.5; const env = envToOneHot(c.environment_preference); v[7] = env[0]; v[8] = env[1]; v[9] = env[2]; v[10] = c.comfortable_with_pets ? 1 : 0; v[11] = c.needs_sibling_placement ? 1 : 0; const langs: string[] = [c.primary_language, ...(c.additional_languages ?? [])].filter(Boolean); v[12] = langs.length > 1 ? 0.8 : 0.5; v[13] = c.cultural_background ? 0.7 : 0.3; v[14] = (c.cultural_practices?.length ?? 0) > 0 ? 0.7 : 0.3; v[15] = c.adaptability ?? 0.5; v[16] = 0.5; v[17] = 0.5; v[18] = c.therapy_ongoing ? 0.7 : 0.3; v[19] = c.verbal_ability ?? 0.5; v[20] = c.expressiveness ?? 0.5; v[21] = (c.special_needs?.length ?? 0) > 0 ? 0.8 : 0.2; v[22] = (c.interests?.length ?? 0) / 10; v[23] = (c.coping_strategies?.length ?? 0) / 5; v[24] = c.adaptability ?? 0.5; return v; } function familyToVector(f: Record): number[] { const v = new Array(25).fill(0); let familyEnergy = 0.5; const activities: string[] = f.family_activities ?? []; if (hasKeyword(activities, ["sport", "outdoor", "hiking"])) familyEnergy += 0.2; if ((f.structure_level ?? 0.5) > 0.7) familyEnergy -= 0.1; v[0] = Math.max(0, Math.min(1, familyEnergy)); v[1] = (f.emotional_availability ?? 0.5) * 0.4 + commStyleToWarmth(f.communication_style) * 0.6; v[2] = f.structure_level ?? 0.5; v[3] = (f.emotional_regulation_capacity ?? 0.5) * 0.4 + (f.patience_level ?? 0.5) * 0.3 + (f.resilience_level ?? 0.5) * 0.3; v[4] = approachToNumeric(f.attachment_approach); v[5] = (f.resilience_level ?? 0.5) * 0.5 + (f.has_trauma_training ? 0.5 : 0); v[6] = (f.patience_level ?? 0.5) * 0.6 + (f.flexibility_level ?? 0.5) * 0.4; const env = envToOneHot(f.home_environment); v[7] = env[0]; v[8] = env[1]; v[9] = env[2]; v[10] = (f.pets?.length ?? 0) > 0 ? 1 : 0; v[11] = f.open_to_sibling_group ? 1 : 0; const langs: string[] = [f.primary_language, ...(f.additional_languages ?? [])].filter(Boolean); v[12] = langs.length > 1 ? 0.8 : 0.5; v[13] = f.cultural_background ? 0.7 : 0.3; v[14] = (f.cultural_practices?.length ?? 0) > 0 ? 0.7 : 0.3; v[15] = f.resilience_level ?? 0.5; v[16] = f.support_network_strength ?? 0.5; v[17] = (f.has_parenting_experience ? 0.3 : 0) + (f.has_foster_experience ? 0.3 : 0) + (f.has_adoption_experience ? 0.2 : 0) + (f.has_trauma_training ? 0.2 : 0); v[18] = f.available_time ?? 0.5; v[19] = f.conflict_resolution_skill ?? 0.5; v[20] = f.emotional_availability ?? 0.5; v[21] = f.open_to_special_needs ? 0.8 : 0.2; v[22] = activities.length / 10; v[23] = (f.completed_trainings?.length ?? 0) / 5; v[24] = f.flexibility_level ?? 0.5; return v; } // ---------- Main Handler ---------- serve(async (req: Request) => { try { const { type, id } = await req.json(); if (!type || !id) { return new Response(JSON.stringify({ error: "type and id required" }), { status: 400 }); } const supabase = createClient(SUPABASE_URL, SUPABASE_SERVICE_KEY); const table = type === "child" ? "children" : "families"; const { data: profile, error: fetchErr } = await supabase .from(table) .select("*") .eq("id", id) .single(); if (fetchErr || !profile) { return new Response(JSON.stringify({ error: fetchErr?.message ?? "Not found" }), { status: 404, }); } const embedding = type === "child" ? childToVector(profile) : familyToVector(profile); const { error: updateErr } = await supabase .from(table) .update({ embedding: `[${embedding.join(",")}]` }) .eq("id", id); if (updateErr) { return new Response(JSON.stringify({ error: updateErr.message }), { status: 500 }); } return new Response(JSON.stringify({ ok: true, embedding }), { headers: { "Content-Type": "application/json" }, }); } catch (e) { return new Response(JSON.stringify({ error: (e as Error [truncated — 40 more characters] ``` ### supabase/functions/extract-caseworker-metrics/index.ts ```typescript import { serve } from "https://deno.land/std@0.177.0/http/server.ts"; import { createClient } from "https://esm.sh/@supabase/supabase-js@2"; const SUPABASE_URL = Deno.env.get("SUPABASE_URL")!; const SUPABASE_SERVICE_KEY = Deno.env.get("SUPABASE_SERVICE_ROLE_KEY")!; const ANTHROPIC_API_KEY = Deno.env.get("ANTHROPIC_API_KEY")!; const SYSTEM_PROMPT = `You are a structured data extraction system for a child adoption matching platform. You will receive free-text caseworker/social worker notes about a child in the foster care or adoption system. Your job is to extract structured metrics from these notes that will be used to match the child with a compatible family. IMPORTANT RULES: - Only extract what the notes explicitly state or strongly imply. Do NOT guess. - If the notes do not mention a field, omit it entirely. - Never store identifying info (last names, addresses, school names, phone numbers). - Use the exact field names and value formats below. EXTRACTED DATA SCHEMA: You MAY include any of these fields when confident: Core fields: - "interests": ["..."] - "energyLevel": "high" | "moderate" | "low" - "socialComfort": "high" | "moderate" | "low" - "emotionalTriggers": ["..."] - "copingStrategies": ["..."] - "trustLevel": "low" | "low-moderate" | "moderate-high" | "high" - "expressiveness": "high" | "moderate" | "low" - "structureNeed": "high" | "moderate" | "low" - "environmentPreference": "urban" | "suburban" | "rural" | "no_preference" - "petComfort": true | false - "siblingNeed": true | false - "communicationStyle": "verbal" | "non-verbal" | "mixed" - "adaptability": "high" | "moderate" | "low" Holistic matching signals: - "strengths": ["..."] - "calmingSupports": ["..."] - "comfortBoundaries": { "touch": "likes_hugs" | "ask_first" | "no_touch" | "depends", "privacy": "needs_space" | "ok_close" | "mixed" } - "routinePreferences": { "morning": "slow" | "fast" | "mixed", "bedtime": "structured" | "flexible" | "mixed" } - "sensoryPreferences": { "noise": "quiet" | "ok_noise" | "depends", "crowds": "avoid" | "ok" | "depends" } - "relationshipPriorities": { "importantPeopleRoles": ["sibling", "grandparent", "mentor", "foster_parent", "friend", "teacher"], "contactImportance": "low" | "medium" | "high" } - "conflictRepairPreference": "space_then_talk" | "talk_now" | "write_text" | "adult_apology_first" | "not_sure" - "transitionSupports": ["preview_photos", "clear_rules", "slow_introductions", "keep_items", "goodbye_rituals"] - "identityNeeds": ["language", "hair_skin_care", "faith", "food_traditions", "community_connection"] - "serviceComfort": { "talking_to_helper": "comfortable" | "maybe" | "not_now" | "depends", "what_helped_before": ["..."] } Needs snapshot (0-3) - ONLY if notes provide clear evidence: - "needsScores": { "emotionalRegulation": 0|1|2|3, "traumaSymptomLoad": 0|1|2|3, "behaviorFunctioning": 0|1|2|3, "developmentalComplexity": 0|1|2|3, "attachmentTransitionSupport": 0|1|2|3, "transitionLossLoad": 0|1|2|3, "identityContinuityNeeds": 0|1|2|3, "serviceIntensityNeeds": 0|1|2|3 } Scoring: 0 = no signal; 1 = mild; 2 = moderate; 3 = intense/safety-relevant. Safety flags (TRUE ONLY if notes explicitly mention): - "safetyFlags": { "currentDanger": true|false, "abuseDisclosure": true|false, "selfHarmIdeation": true|false, "harmToOthersIdeation": true|false, "runningAwayRisk": true|false, "exploitationRisk": true|false } Caseworker-specific fields (new - only from professional observation): - "attachmentStyle": "secure" | "anxious" | "avoidant" | "disorganized" - "traumaSeverity": 0.0 to 1.0 - "emotionalAwareness": 0.0 to 1.0 - "verbalAbility": 0.0 to 1.0 - "therapeuticNeeds": ["..."] - "placementHistory": { "count": N, "longestStay": "duration", "disruptions": N } - "diagnosisNotes": ["..."] OUTPUT: Respond with ONLY valid JSON (no markdown, no code fences): { "extractedMetrics": { ... }, "confidence": "high" | "medium" | "low", "summary": "1-2 sentence summary of what you extracted" }`; serve(async (req: Request) => { try { const { childName, childAge, notes, childId } = await req.json(); if (!childName || !notes) { return new Response( JSON.stringify({ error: "childName and notes are required" }), { status: 400 } ); } // Call Claude API const apiResponse = await fetch("https://api.anthropic.com/v1/messages", { method: "POST", headers: { "Content-Type": "application/json", "x-api-key": ANTHROPIC_API_KEY, "anthropic-version": "2023-06-01", }, body: JSON.stringify({ model: "claude-sonnet-4-5-20250929", max_tokens: 2048, system: SYSTEM_PROMPT, messages: [ { role: "user", content: `Extract structured metrics from the following caseworker notes about ${childName} (age ${childAge ?? "unknown"}):\n\n${notes}`, }, ], }), }); if (!apiResponse.ok) { const errBody = await apiResponse.json(); return new Response( JSON.stringify({ extractedMetrics: {}, confidence: "low", summary: "Failed to process notes.", error: `API Error ${apiResponse.status}: ${errBody?.error?.message ?? "Unknown"}`, }), { status: 200, headers: { "Content-Type": "application/json" } } ); } const body = await apiResponse.json(); const content = body.content?.[0]?.text ?? "{}"; // Parse response - strip markdown fences if present let cleaned = content.trim(); if (cleaned.startsWith("```json")) cleaned = cleaned.slice(7); else if (cleaned.startsWith("```")) cleaned = cleaned.slice(3); if (cleaned.endsWith("```")) cleaned = cleaned.slice(0, -3); cleaned = cleaned.trim(); let parsed: Record; try { parsed = JSON.parse(cleaned); } catch { parsed = { extra [truncated — 1287 more characters] ``` ### supabase/functions/compute-matches/index.ts ```typescript import { serve } from "https://deno.land/std@0.177.0/http/server.ts"; import { createClient, SupabaseClient } from "https://esm.sh/@supabase/supabase-js@2"; const SUPABASE_URL = Deno.env.get("SUPABASE_URL")!; const SUPABASE_SERVICE_KEY = Deno.env.get("SUPABASE_SERVICE_ROLE_KEY")!; const ALGORITHM_VERSION = "2.0.0-multilayer"; const DIMENSION_WEIGHTS: Record = { personality: 0.15, emotional: 0.2, attachment: 0.2, cultural: 0.1, communication: 0.1, environment: 0.1, stability: 0.15, }; // ── Types ── interface MatchDimension { name: string; score: number; weight: number; explanation: string; child_factors: string[]; family_factors: string[]; } interface MatchFactor { factor_name: string; impact: number; explanation: string; category: string; } interface ReasoningLayer { layer_number: number; name: string; description: string; score_adjustment?: number; score_after?: number; findings: string[]; reasoning: string[]; data: Record; passed: boolean; } interface MatchResult { id: string; child_id: string; family_id: string; child_name: string; family_name: string; overall_score: number; personality_score: number; emotional_score: number; attachment_score: number; cultural_score: number; communication_score: number; environment_score: number; stability_score: number; dimensions: Record; strengths: string[]; challenges: string[]; recommendations: string[]; bias_flags: string[]; placement_stability_prediction: number; risk_level: string; risk_factors: string[]; protective_factors: string[]; scoring_factors: MatchFactor[]; algorithm_version: string; generated_date: string; reasoning_layers: ReasoningLayer[]; status: string; } type Child = Record; type Family = Record; type Transcript = Record; type CWNotes = Record; // ── Utility ── function clamp(v: number, lo: number, hi: number): number { return Math.max(lo, Math.min(hi, v)); } function levelLabel(v: number): string { if (v < 0.25) return "Low"; if (v < 0.5) return "Moderate"; if (v < 0.75) return "High"; return "Very High"; } function calculateOverlap(a: string[], b: string[]): number { if (!a?.length || !b?.length) return 0.3; const s1 = new Set(a.map((x) => x.toLowerCase())); const s2 = new Set(b.map((x) => x.toLowerCase())); let inter = 0; s1.forEach((x) => { if (s2.has(x)) inter++; }); const union = new Set([...s1, ...s2]).size; return union > 0 ? inter / union : 0; } // ── Layer 1: Profile Normalization ── function layer1Normalize(child: Child, family: Family): ReasoningLayer { const findings: string[] = []; const reasoning: string[] = []; const data: Record = {}; const childNeedLevel = (child.trauma_severity ?? 0) * 0.3 + (1 - (child.trust_level ?? 0.5)) * 0.25 + (1 - (child.emotional_awareness ?? 0.5)) * 0.2 + (child.structure_need ?? 0.5) * 0.15 + ((child.special_needs?.length ?? 0) > 0 ? 0.1 : 0); data.childNeedLevel = childNeedLevel; const familyCapacity = (family.patience_level ?? 0.5) * 0.2 + (family.emotional_regulation_capacity ?? 0.5) * 0.2 + (family.resilience_level ?? 0.5) * 0.15 + (family.support_network_strength ?? 0.5) * 0.15 + (family.available_time ?? 0.5) * 0.15 + (family.has_trauma_training ? 0.15 : 0); data.familyCapacity = familyCapacity; const capacityGap = familyCapacity - childNeedLevel; data.capacityGap = capacityGap; let familyEnergy = 0.5; const acts: string[] = family.family_activities ?? []; if (acts.some((a: string) => /sport|outdoor|hiking/i.test(a))) familyEnergy += 0.2; if ((family.structure_level ?? 0.5) > 0.7) familyEnergy -= 0.1; data.familyEnergy = clamp(familyEnergy, 0, 1); findings.push(`Child need level: ${(childNeedLevel * 100).toFixed(0)}%`); findings.push(`Family capacity: ${(familyCapacity * 100).toFixed(0)}%`); findings.push(`Capacity gap: ${capacityGap > 0 ? "+" : ""}${(capacityGap * 100).toFixed(0)}%`); reasoning.push( `Extracted ${(child.interests?.length ?? 0) + (child.coping_strategies?.length ?? 0) + (child.emotional_triggers?.length ?? 0)} child feature signals` ); reasoning.push( `Extracted ${(family.family_activities?.length ?? 0) + (family.completed_trainings?.length ?? 0) + (family.family_values?.length ?? 0)} family feature signals` ); if (capacityGap > 0.1) reasoning.push("Family capacity exceeds child needs - positive indicator"); else if (capacityGap < -0.1) reasoning.push("Child needs may exceed family capacity - requires careful evaluation"); else reasoning.push("Family capacity and child needs are closely matched"); return { layer_number: 1, name: "Profile Normalization", description: "Extracts composite features from raw profile data and normalizes them for cross-comparison.", findings, reasoning, data, passed: true, }; } // ── Layer 2: Hard Constraints ── function layer2HardConstraints(child: Child, family: Family): ReasoningLayer { const findings: string[] = []; const reasoning: string[] = []; let penalty = 0; let passed = true; if (child.age < (family.min_child_age ?? 0) || child.age > (family.max_child_age ?? 18)) { findings.push(`CONSTRAINT: Child age ${child.age} outside family range ${family.min_child_age}-${family.max_child_age}`); reasoning.push("Age mismatch is a soft constraint - penalizes score by 10 points"); penalty += 10; } else { findings.push(`Age check passed: ${child.age} within ${family.min_child_age}-${family.max_child_age}`); } if (child.needs_sibling_placement && !family.open_to_sibling_group) { findings.push("CONSTRAINT: Child requires sibling placement, family not open to sibling groups"); reasoning.push("Sibling separation is a high-risk factor - penalizes score by 15 points"); penalty += 15; } else if (chil [truncated — 38244 more characters] ``` ### analysis_options.yaml ```yaml include: package:flutter_lints/flutter.yaml linter: rules: prefer_const_constructors: false prefer_const_literals_to_create_immutables: false avoid_print: false ``` ### run_dev.sh ```shell #!/bin/bash # Launch Flutter Web with CORS disabled for development. # HeyGen's Streaming Avatar REST API doesn't allow browser-origin requests, # so we need to disable web security in Chrome for the API calls to work. # # Usage: ./run_dev.sh flutter run -d chrome \ --web-browser-flag="--disable-web-security" \ --web-browser-flag="--user-data-dir=/tmp/flutter_chrome_dev" ``` ### pubspec.yaml ```yaml name: adoption_matching description: AI-powered adoption matching system for improving orphan-family pairing. publish_to: 'none' version: 1.0.0+1 environment: sdk: '>=3.0.0 <4.0.0' dependencies: flutter: sdk: flutter cupertino_icons: ^1.0.6 fl_chart: ^0.68.0 google_fonts: ^6.1.0 percent_indicator: ^4.2.3 uuid: ^4.2.1 provider: ^6.1.1 http: ^1.2.0 supabase_flutter: ^2.3.0 flutter_tts: ^4.2.0 speech_to_text: ^7.0.0 web: ^1.1.0 dev_dependencies: flutter_test: sdk: flutter flutter_lints: ^3.0.1 flutter: uses-material-design: true ``` ### web/index.html ```html The First Match ``` ### lib/main.dart ```dart import 'package:flutter/material.dart'; import 'package:provider/provider.dart'; import 'package:supabase_flutter/supabase_flutter.dart'; import 'config/secrets.dart'; import 'services/app_state.dart'; import 'utils/theme.dart'; import 'screens/home_screen.dart'; import 'screens/dashboard_screen.dart'; import 'screens/child_onboarding_screen.dart'; import 'screens/family_onboarding_screen.dart'; import 'screens/matches_screen.dart'; import 'screens/match_detail_screen.dart'; import 'screens/children_screen.dart'; import 'screens/families_screen.dart'; import 'screens/counselor_chat_screen.dart'; import 'screens/caseworker_notes_screen.dart'; void main() async { WidgetsFlutterBinding.ensureInitialized(); await Supabase.initialize( url: Secrets.supabaseUrl, anonKey: Secrets.supabaseAnonKey, ); runApp( ChangeNotifierProvider( create: (_) => AppState(Supabase.instance.client), child: const AdoptMatchApp(), ), ); } class AdoptMatchApp extends StatelessWidget { const AdoptMatchApp({super.key}); @override Widget build(BuildContext context) { return MaterialApp( title: 'The First Match', debugShowCheckedModeBanner: false, theme: AppTheme.theme, initialRoute: '/', routes: { '/': (context) => const HomeScreen(), '/dashboard': (context) => const DashboardScreen(), '/child-onboarding': (context) => const ChildOnboardingScreen(), '/family-onboarding': (context) => const FamilyOnboardingScreen(), '/matches': (context) => const MatchesScreen(), '/match-detail': (context) => const MatchDetailScreen(), '/children': (context) => const ChildrenScreen(), '/families': (context) => const FamiliesScreen(), '/counselor-chat': (context) => const CounselorChatScreen(), '/caseworker-notes': (context) => const CaseworkerNotesScreen(), }, ); } } ``` ### web/heygen_livekit_bridge.js ```javascript /** * HeyGen LiveKit Bridge for Flutter Web * * This script connects to a HeyGen Streaming Avatar session via LiveKit WebRTC * and attaches the remote video track to a