AISaaSNext.jsTypeScript

Thread.ai

AI knowledge co-pilot for engineering teams. Indexes codebases, Slack history, docs, and PRs into a queryable knowledge graph — so developers spend less time hunting context and more time shipping.

screenshot

+25%

Productivity gain

-30%

Cost reduction

500+

Active users

The Problem

Every engineering team carries an invisible tax: context loss. New hires spend weeks reading stale docs. Experienced engineers re-answer the same questions in Slack. Critical architectural decisions live in 2-year-old threads nobody can find.

The average developer spends 23 minutes finding context for a single question. Multiply that across a 20-person team and you've lost days of productivity every week.

The Solution

Thread.ai ingests a team's entire knowledge surface — Slack history, GitHub PRs, Notion pages, Jira tickets, internal docs — and makes it instantly queryable in natural language.

Ask "why did we migrate away from MongoDB?" and get an answer with links to the exact Slack thread, PR, and ADR where that decision was made.

Technical Approach

The core system is a RAG (Retrieval-Augmented Generation) pipeline:

  1. Ingestion — Documents are chunked with a sliding window (512 tokens, 64 overlap) and embedded via OpenAI's text-embedding-3-small
  2. Storage — Embeddings stored in PostgreSQL with pgvector extension for fast cosine similarity search
  3. Retrieval — On query, the user's question is embedded and top-k chunks retrieved with a minimum similarity threshold of 0.78
  4. Generation — Retrieved context fed to GPT-4 with strict source citation requirements

The strict similarity threshold was the key reliability unlock — "I don't know" is far better than a confident hallucination.

Results

After six weeks in beta with four enterprise teams:

Tech stack

Next.jsFrontend
Nest.jsBackend
TypeScriptLanguage
PostgreSQL + pgvectorDatabase
OpenAI APIAI
RedisCache
AWSInfrastructure

My role

Lead Engineer — owned the full stack from ingestion pipeline to chat UI. Led architecture decisions, built the RAG pipeline, shipped the beta, and iterated based on user feedback.

Got something similar to build?

I'm available for select freelance work — AI products, full-stack SaaS, architecture.

Let's talk
Back to all projects