AskTiz

Internal AI Knowledge Assistant

What we built

Developed pipelines that continuously ingest Slack messages using the Slack Events API.
Built asynchronous processing queues using PostgreSQL pgmq for reliable message ingestion.
Automated embedding generation and indexing for newly received conversations.
Supported incremental synchronization without interrupting existing search indexes.

Designed and implemented a complete RAG pipeline for enterprise knowledge retrieval.
Combined semantic retrieval with lexical search to maximize answer accuracy.
Enabled contextual AI responses grounded entirely in company conversations.
Returned inline citations linking directly to original Slack messages for transparency and verification.

Implemented hybrid retrieval combining PostgreSQL full-text search with pgvector semantic similarity search.
Leveraged Voyage AI embeddings for high-quality semantic representations.
Integrated Cohere Rerank to reorder search results and improve relevance before passing context to the LLM.
Significantly improved retrieval performance for acronyms, product names, technical terminology, and company-specific vocabulary.

Utilized Claude Haiku to intelligently rewrite user queries and improve retrieval recall.
Generated final responses using Claude Sonnet with citation-aware prompting.
Built streaming AI responses using Server-Sent Events (SSE) for real-time conversational experiences.
Implemented graceful fallback strategies for long-running AI requests.

Designed a backend-first architecture where all authentication, authorization, and database access remain within the Fastify API.
Prevented direct database access from the frontend, reducing the application's security surface.
Built dedicated authentication and chat APIs consumed by the Next.js application.

Developed reproducible PostgreSQL schemas using raw SQL migrations.
Configured pgvector indexes and PostgreSQL full-text search infrastructure.
Implemented Row-Level Security (RLS) to ensure users can only retrieve knowledge from Slack channels they are authorized to access.
Built scalable indexing strategies optimized for large conversation histories.

Developed a modern Next.js 14 interface for conversational knowledge retrieval.
Implemented real-time streaming responses with inline citations and source navigation.
Built an intuitive chat interface supporting continuous conversations and contextual follow-up questions.

Real-time Slack Events API ingestion with asynchronous queue-based processing.
Hybrid search architecture combining semantic search, lexical search, and AI-powered reranking.
Enterprise-grade RAG pipeline delivering highly accurate, citation-backed responses.
Secure backend-first architecture with centralized authentication and authorization.
Streaming conversational AI powered by Claude with low-latency user experiences.
PostgreSQL optimized with pgvector, pgmq, full-text search, and Row-Level Security.

What it does