Insightio
AI user-research synthesis for product teams
Visit Insightio → https://insightio.webflow.io/The Problem
- Feedback is scattered across reviews, interviews, surveys, support, and success notes, which makes it slow to synthesize, hard to trust, and easy to miss patterns.
What We Built
Multimodal Ingestion
Text (CSV/Docs), audio, and video into a single workspace.
Transcription on AWS
Audio/video routed to AWS Transcribe; artifacts stored in Amazon S3 and referenced in-app.
Normalization Pipeline
De-duplication, language detection, PII scrubbing, speaker attribution for interviews.
Thematic Clustering
Vector index + topic modeling to surface recurring themes, sentiments, intents, JTBD.
Insight Cards
Summaries with source links, timestamps, and confidence for traceability from insight → sentence → source.
NLQ Search
Ask natural questions (e.g., "Top onboarding blockers for new users?") and get evidence-backed answers.
Prioritization
ICE/RICE fields to rank pain points and opportunities.
Handoffs
One-click exports to PRD/Slack/Notion.
Semantic Chat
One of the best chats to be created back then, helping users expand on insights and problem spaces.
System/Infra Highlights
- Web app frontend connected to a backend orchestrator.
- Parallel API orchestration for ingestion, transcription, embedding, and clustering to cut end-to-end latency.
- S3 as the single source of truth for raw files + transcripts; metadata stored alongside embeddings for fast retrieval.
How It Works
- Upload/import text, audio, or video → store in S3.
- For A/V: run AWS Transcribe → attach transcripts + timestamps.
- Clean & normalize → embed → cluster themes & sentiments using AI.
- Generate Insight Cards with direct links back to sources.
Outcomes
- Compressed weeks of manual synthesis into same-day insight cycles.
- Higher signal, lower bias via consistent clustering and evidence-backed summaries.
- A shared, traceable source of truth across PM, Design, and CS.