TalkPick

Take a look

Turn chat exports into AI reports — a multi-platform conversation analysis SaaS

Client: Self-funded product
Industry: Consumer · AI / Data
Year: 2025 — Live
Scope: Product · UX / UI design · Full-stack development · Infrastructure · Operations
Stack: Next.js 15FastAPIPostgreSQLMongoDBChromaDBOpenAIAWS SES/Lambda/S3PortOne

Overview

A B2C SaaS that turns chat exports from KakaoTalk, Instagram, Telegram, LINE, and WhatsApp into AI-generated reports — emotional flow, relationship diagnosis, and conversation habits. Ships with a RAG-based chatbot, multi-section reports, sharing, and payments — all production-ready.

Challenge

Chat data is rich, but tools that turn it into meaningful insight are rare. The bigger problem is friction — KakaoTalk only exports chats on mobile, and getting that zip onto a desktop browser (or uploading it directly from a mobile browser) is where most users drop off.

Solution

Built per-platform parsers, vector embeddings, and an LLM analysis pipeline — and made the upload path itself the simplest possible: send the file as an email attachment.

Multi-platform parsers — auto-detect and parse 5 SNS formats
RAG chatbot — semantic chunking with natural-language Q&A
12+ section AI report — emotion charts, relationship diagnosis, best moments, couple's dictionary
Sharing + OG image — public links, KakaoTalk preview cards
Payments — subscription (Basic / Super) + single-report purchase via PortOne

Signature

Email-only upload — AWS SES × Lambda

Send mail to upload+<token>@email.talk-pick.com and the AWS SES → S3 → Lambda → Backend pipeline picks up the attachment, auto-merges KakaoTalk split zips, decodes MIME with EUC-KR fallback, and replies via SES — users get a result notification before they ever come back to the app.

"The shortest possible upload path — no desktop, no app switch."

Result

Live · paid plans (subscription + single-purchase)
Korean / English support
Sentry · CloudWatch monitoring in production

Gallery