Taledge

Talent Intelligence & Success Platform prototype. A dual-track web app that fuses DNLA psychometrics, AI-driven interview evaluation, and human coaching into a closed-loop assessment journey for candidates, institutes, recruiters, and coaches.

Stack

• Next.js 15 (App Router) + React 19
• Tailwind CSS 3
• Google Gemini API
  • gemini-2.5-flash for resume parsing, interview turns, and Fit Score synthesis
  • gemini-3.1-flash-live-preview for Gemini Live token provisioning
• pnpm

Local development

pnpm install
cp .env.example .env.local   # then set GEMINI_API_KEY
PORT=4040 pnpm dev

Open http://localhost:4040.

Environment

Create a .env.local file in the project root (copy from below or create manually):

# Required: Gemini API key for all AI features
GEMINI_API_KEY=your_gemini_api_key_here
GEMINI_TEXT_MODEL=gemini-2.5-flash
GEMINI_LIVE_MODEL=gemini-3.1-flash-live-preview

Getting Your Gemini API Key

1. Go to Google AI Studio
2. Click "Create API Key"
3. Copy the key (starts with AIza...)
4. Paste it in your .env.local file as shown above

Note: This repo is Gemini-only. It does not require third-party LLM router or non-Google provider packages.

Variable	Purpose
GEMINI_API_KEY	Required for resume parsing, interview questions, Fit Score synthesis, and Gemini Live token provisioning. Without it, AI endpoints return 503 instead of fake results.
GEMINI_TEXT_MODEL	Text/multimodal Gemini model. Defaults to gemini-2.5-flash.
GEMINI_LIVE_MODEL	Live API model. Defaults to gemini-3.1-flash-live-preview.

Voice Interview Pipeline

The Phase 1 interview uses a text fallback and browser speech input while Gemini Live credentials are being provisioned. /api/gemini/live-token mints short-lived tokens for the Live API once GEMINI_API_KEY is present.

Flow

1. Start session → POST /api/interview/start
2. Send text answer → POST /api/interview/voice
3. Receive the next question
4. Repeat until isDone: true
5. Get final results → POST /api/interview/results

Audio Format

• Browser speech recognition can fill the text answer field.
• Server audio upload is intentionally disabled; use Gemini Live with ephemeral tokens for production audio sessions.

API

All /api/interview/* routes require studentId, role, and mode.

• POST /api/interview — multi-turn interview question generation (text-based, existing)
• POST /api/interview/start — start a voice session, returns sessionId
• POST /api/interview/voice — receive a text answer and return the next question
• POST /api/gemini/live-token — mint a Gemini Live ephemeral token
• GET /api/interview/status?sessionId=... — get current session state
• POST /api/interview/results — get final scores (only after isDone: true)
• POST /api/parse-resume — PDF resume parsing
• POST /api/generate-dnla — disabled until DNLA provider import is connected
• POST /api/generate-fit-score — Final Fit Score report synthesis

Candidate flow (no top nav, focused experience):

• /onboarding — Step 01 Profile + Goal + Context (PDF resume parsed by Gemini 2.5 Flash)
• /student/[id] — Candidate workspace
• /student/[id]/interview/technical — Step 02 AI Technical Interview (Gemini 2.5 Flash)
• /student/[id]/dnla — Step 03 DNLA import placeholder
• /student/[id]/interview/behavioural — Step 04 AI Behavioural Interview (Gemini 2.5 Flash)
• /student/[id]/fit-score — Step 05 Fit Score Reveal (Gemini-generated, PRD §9 rubric)
• /student/[id]/development — Step 06 Development Pathway

Marketing and B2B surfaces (with top nav):

• / — Landing (ClaimKeys-style sign-in split layout)
• /exam/[id] — Competitive Exam aspirant
• /institute/[id] — Placement / Exam institute dashboard
• /recruiter/[id] — Recruiter console with filtering
• /coach/[id] — Coach workspace
• /coach-ai — hidden Phase 2 route with unavailable screen

API:

• POST /api/interview — multi-turn interview question generation
• POST /api/parse-resume — PDF resume parsing
• POST /api/generate-dnla — disabled until DNLA provider import is connected
• POST /api/generate-fit-score — Final Fit Score report synthesis

Smoke Test — Voice Pipeline

1. Start session

curl -X POST http://localhost:3000/api/interview/start \
  -H "Content-Type: application/json" \
  -d '{"studentId":"priya","role":"Full-stack Software Engineer","mode":"technical"}'
# → Returns: { "ok": true, "sessionId": "session_...", "mode": "technical" }

2. Send first text answer

curl -X POST http://localhost:3000/api/interview/voice \
  -H "Content-Type: application/json" \
  -d '{"sessionId":"<session_id>","text":"I built a full-stack app using React and Node.js"}'
# → Returns: { "ok": true, "nextQuestion": "...", "isDone": false }

3. Mint a Gemini Live token

curl -X POST http://localhost:3000/api/gemini/live-token
# → Returns 503 without GEMINI_API_KEY, or a short-lived token when configured

4. Check session state

curl "http://localhost:3000/api/interview/status?sessionId=<session_id>"
# → Returns: { "ok": true, "turnIndex": 1, "transcript": [...], ... }

5. End interview & get results

Send enough responses until isDone: true, then:

curl -X POST http://localhost:3000/api/interview/results \
  -H "Content-Type: application/json" \
  -d '{"sessionId":"<session_id>"}'
# → Returns: { "ok": true, "generated": { "technical_score": 72, "fit_score": 70, ... } }

lib/data.ts currently provides anonymized local workspace shells only. Production persistence, auth, and organization-specific records should be connected before real users are onboarded.

Scripts

pnpm dev      # next dev
pnpm build    # next build
pnpm start    # next start