Full-time · Remote (India)

AI Memory & Knowledge Graph Engineer

Cehpoint is hiring for this role: full-time, Remote (India), ₹11,00,000 – ₹19,00,000 / year. Design persistent episodic memory, hierarchical semantic caches, GraphRAG engines, and cross-session entity reconciliation for autonomous enterprise agents.

About the role

Autonomous agents without persistent, indexed memory are amnesiac. You will build Cehpoint's central memory substrate—combining temporal episodic logs, semantic entity graphs, and dense vector embeddings into a cohesive, sub-100ms retrieval engine that powers all production agents.

We are seeking an AI Memory Engineer to solve long-horizon state persistence and multi-hop associative recall. You will engineer vector-graph hybrid indexes that allow agents to retain knowledge across millions of corporate interactions without context degradation.

What you will do

  • Architect hybrid vector-graph retrieval pipelines combining pgvector, Qdrant, and Neo4j for multi-hop relationship queries.
  • Implement self-consolidating memory algorithms that distill raw conversational turns into structured knowledge graph assertions.
  • Develop real-time entity resolution to resolve ambiguous references across corporate documents and communication channels.
  • Engineer memory decay, importance weighting, and semantic pruning pipelines to maintain sub-second context retrieval.
  • Ensure rigorous tenant isolation and cryptographic access control across multi-user memory partitions.

What you should know

  • Deep theoretical foundation in associative memory models, vector indexing (HNSW, IVF), and graph theory.
  • Understanding of modern RAG paradigms: Dense Retrieval, ColBERT multi-vector representation, and Graph-augmented generation.
  • Database internals: PostgreSQL query planner, write-ahead logs, and transaction isolation levels.

What you should be able to do

  • Extensive production experience with Python (FastAPI/AsyncIO) and/or Node.js.
  • Hands-on expertise with vector databases (pgvector, Qdrant, Pinecone) and graph systems (Neo4j, NetworkX).
  • Proficiency in designing automated data extraction pipelines with schema-validated JSON outputs.

How you work

  • Deep fascination with human hippocampal memory consolidation and cognitive architectures.
  • Uncompromising focus on zero-loss recall and deterministic reproducibility.
  • Passion for benchmarking retrieval latency under high-concurrency production load.

Skills we check

  • Episodic Memory Systems
  • GraphRAG & Neo4j
  • Dense Vector Retrieval
  • pgvector / Qdrant
  • Context Compaction
  • Python / AsyncIO
  • PostgreSQL
  • Knowledge Graphs

What you get

  • Annual package of ₹11,00,000–₹19,00,000 with annual technical appraisal.
  • Ownership of critical core platform IP utilized across all commercial clients.
  • Flexible remote culture with flexible hours and high technical autonomy.
  • Hardware stipend for workstation, monitors, and local development lab.

How we select for this role

We read every application ourselves. Here is what happens after you apply.

  1. Apply with a short form

    Paste a link to your CV and answer five questions in your own words. No sign-in and no file upload. About fifteen minutes.

  2. Integrity check

    When you press Check my answers, an automated check confirms the answers were typed by you. You see the result straight away, before you submit.

  3. Choose your project date

    If no cheating is detected you are shortlisted, and we email you a link to choose the day your project starts.

  4. 5-day practical project

    On that day our team contacts you with the brief: real work, 5 working days, reviewed by an engineer. For this role: Implement a continuous episodic memory indexing engine that compresses and retrieves cross-session context across 1,000 historical agent interactions with sub-100ms retrieval and zero hallucinated recall (strictly 5 working days).

  5. Offer

    If your work solves the problem well, you receive your offer letter by email and in the portal.