Chandan Rajah
Chandan Rajah is an independent technologist, entrepreneur and open-source creator focused on deep tech, AI and data systems. His current open-source work includes post-graph-rag, a PostgreSQL-native Graph RAG engine for retrieving changing, temporally grounded knowledge, and post-graph, the graph-storage layer beneath it.
At a Glance
- Developers and engineering teams already operating PostgreSQL
- AI/ML teams building Graph RAG, retrieval and agent-memory systems
- Organizations processing temporally changing enterprise documents
- Teams requiring tenant isolation, auditability and application/database transaction consistency
AI Tools by Chandan Rajah
(1)
post-graph-rag
Graph RAG for PostgreSQL
Discussions
No discussions yet
Be the first to start a discussion about Chandan Rajah
Latest News
post-graph-rag requires post-graph 1.6.0; post-graph adds explicit replace=True payload writes
Documentation update moved installation, the CFO supersession example and project links above the fold
Release 1.12.0 adds belief-time document filtering and GraphRAG.watermark()
Release 1.11.1 removes the default context cap and reports improved LongMemEval results
Products & Services
Open-source Python Graph RAG library distributed through GitHub and PyPI. It extracts entities and relations with an LLM, stores document chunks, embeddings and a canonical property graph in PostgreSQL with pgvector, and answers questions through vector similarity, graph traversal, full-text search and community summaries. Its distinguishing capability is temporal supersession: later documents can close earlier incompatible facts, and queries can be evaluated as of a time.
Standalone PostgreSQL-backed graph database library beneath post-graph-rag. It provides multi-tenant realms, application-level spaces, pgvector search across current and historical rows, audit logging, append-only history, recursive CTE traversals, JSONB range queries and openCypher support.
Market Position
Rajah positions post-graph-rag against Microsoft GraphRAG, LightRAG and Graphiti/Zep. It differentiates on a single PostgreSQL datastore, transactionality with application tables, controlled/queryable predicates and document-order supersession of contradictory facts; the project page notes that Graphiti is also bi-temporal but depends on extracted dates to express when a fact ended. The paper and project materials also discuss HippoRAG and RAPTOR as adjacent Graph RAG/retrieval approaches.
Leadership
Founders
Chandan Rajah
Technologist with more than two decades in deep tech; founder/CEO of Parallel AI, founder of Karedo, co-founder and CTO of LoopCycle, and former VP of Engineering at Noon Academy. Public profiles also describe leadership or engineering experience at Meta, Amazon Web Services/AWS and SambaNova. He holds an MSc in Computer Engineering from BITS Pilani, an MSc in Artificial Intelligence and Robotics from the University of Essex, and an MBA from London Business School.
Founding Story
Rajah describes beginning his Web3 journey in 2013 by creating a company intended to tokenize attention and intent, giving consumers ownership on advertising platforms. His later independent work reflects the same deep-tech/AI focus: post-graph-rag was created to address the infrastructure fragmentation, low-quality extraction and lack of temporal reasoning he identifies in conventional Graph RAG systems, while keeping the graph, embeddings and application data in PostgreSQL.
Business Model
Revenue Model
The projects are open-source Apache 2.0 libraries distributed through public GitHub repositories and PyPI. The public materials do not describe a paid hosted service, subscription, API metering or commercial license for these projects.
Target Markets
- Developers and engineering teams already operating PostgreSQL
- AI/ML teams building Graph RAG, retrieval and agent-memory systems
- Organizations processing temporally changing enterprise documents
- Teams requiring tenant isolation, auditability and application/database transaction consistency
- Long-horizon conversational or agent memory where later statements should supersede earlier facts
- Knowledge bases built from changing documents such as contracts, filings, case law, news archives and employment histories
- Financial and earnings-call analysis involving repeated restatements over time
- Applications that need a knowledge graph and embeddings transactionally consistent with PostgreSQL application tables
- Corpus-level thematic question answering using community summaries