Built an early-stage agentic system where Slack and Email agents interpret user intent, query a structured knowledge graph, and return contextual responses. Key contributions:
Designed a standard agent flow: input → intent classification → knowledge graph query → response/hand-off
Implemented Slack and Email agents, routing messages through a shared backend pipeline
Built a Python-based Google Drive metadata crawler to automate asset ingestion (name, type, owner, timestamps) into a structured catalogue
Identified and documented a critical permission bottleneck with Google Drive APIs (metadata access constrained by org-level permissions)
Designed initial knowledge graph schema (assets, tags, relations) to support agent queries like “find latest policy doc”
Hardest challenge: The crawler failed on live data due to restricted Drive permissions. I validated that both Sheets and Drive APIs respect access controls, documented the limitation, and proposed permission-aware ingestion workflows instead of brittle scraping.
Tech stack: Python, HuggingFace transformers, Google APIs, JSON-LD / graph schemas, Slack bot integration