Hindsight

The description states that Hindsight is a tool for reconstructing disputed client invoices across multiple systems (finance, CRM, email, chat, project tools). It builds an ordered evidence trail from authorized records and separates results into four categories: timeline, conflict, gap, and next action.

It uses AI models (e.g., GPT-5.6-sol) to parse questions, compose structured outputs, and answer follow-ups within the bounds of available evidence. Evidence IDs are returned instead of URLs directly, and access is validated server-side before resolving receipts.

The system is built as a TypeScript modular monolith using Next.js, React, PostgreSQL, Inngest, and integrates with services like Clerk for authentication, Nango for connectors, and OpenAI for language processing.

Not evidenced: whether this is a managed SaaS product or still in prototype form; no mention of actual customers, revenue, or deployment beyond the hackathon demo.

The description claims Hindsight helps agency owners enter difficult client conversations with an ordered, verifiable account of what happened. It positions itself as a tool for operations teams to resolve disputes by showing which evidence supports each conclusion and what is missing.

It explicitly avoids declaring fault or assigning blame — the user makes those judgments. The product is framed as a neutral investigation assistant rather than a decision engine.

Inferred: This suggests a positioning shift from generic document search toward structured dispute resolution, but there is no evidence of prior versions or evolution in claims beyond the hackathon context.

The description states that Hindsight targets agency owners and operations leads who deal with client disputes involving invoices, contracts, scope changes, and project deliverables. These users are expected to have access to systems like HubSpot, Slack, Asana, QuickBooks, and Gmail.

Not evidenced: No explicit segmentation or persona definition beyond “operations teams” or “agency owners.” No evidence of customer interviews, usage data, or target market size.

The description does not mention pricing, subscriptions, or monetization strategies. It only describes the product’s functionality and architecture.

Not evidenced: No indication of a business model, pricing tiers, or revenue streams beyond the hackathon demo.

Hindsight is built as a TypeScript modular monolith using Next.js, React, PostgreSQL, Inngest, and integrates with tools like Clerk, Nango, OpenAI, and others. It uses structured outputs (Zod) to manage AI responses and enforces access control at the retrieval level.

The system includes:

• Normalized event data
• ACL-aware retrieval
• Evidence ID-based citations
• Server-side receipt resolution
• Multi-model architecture (GPT for composition, smaller model for entity linking)

Not evidenced: No information on scaling, performance metrics, or production stability. No mention of cloud infrastructure beyond Vercel and Cloudflare.

The project is described as a hackathon submission from the OpenAI 2026 hackathon. It includes a demo at hindsight.questili.com and a repository with local setup instructions.

Not evidenced: No evidence of user adoption, customer feedback, or product traction beyond the demo environment. No mention of funding, headcount, or growth metrics.

The description does not reference direct competitors or similar tools in the market. It implies that existing systems (CRM, email, project management platforms) do not provide a unified view for dispute resolution.

Inferred: Hindsight may compete with or complement tools like HubSpot, Salesforce, Asana, and QuickBooks by offering a structured way to cross-reference data during disputes, but no competitive analysis is provided.

• Unverified claims: All information comes from the author’s own description — no independent validation.
• Demo-only status: No evidence of production use or real-world adoption.
• No business model clarity: No indication of how Hindsight will generate revenue.
• Limited access control testing: While access boundaries are enforced, there is no evidence of robustness in real-world usage.
• AI dependency without transparency: The system relies heavily on AI for structuring outputs; no mention of explainability or auditability features.

1. Has Hindsight been used in production by any clients yet?
2. What is the plan for monetization and pricing?
3. How does Hindsight handle data privacy and compliance (e.g., GDPR, CCPA)?
4. Are there plans to support more integrations beyond those mentioned in the tech stack?
5. What are the key assumptions about user behavior or workflows that underpin this product?
6. How is the AI output validated or corrected by users?
7. What are the main technical challenges encountered during development and deployment?