PROCESS

What I Built

I designed a modular AI assistant that automatically detects learning-related emails, extracts structured event data, evaluates relevance, and adds qualified events directly to my calendar.

Although implemented as a single orchestrated workflow, the system is intentionally structured into distinct reasoning stages, making it easy to extend into a multi-agent system if scaled.

How the System Works (AI in Action)

Stage 1 — Event Detection

Defined clear criteria for what qualifies as an event (workshops, talks, courses, meetups)
Used an LLM to classify incoming emails as event or non-event
Filtered out noise early to reduce downstream errors

Stage 2 — Information Extraction

Designed a fixed schema for event data:
- Title
- Short description
- Date and time
- Format (online / in-person)
- Source
Used prompt constraints to normalize inconsistent email formats into structured outputs

Stage 3 — Relevance Evaluation

Implemented decision logic based on:
- Topic relevance
- My availability (date & time)
- Event format (online vs in-person)
Automatically determined whether an event deserved calendar space
Applied topic and format tags for later filtering

Stage 4 — Action Execution

Connected structured outputs to a calendar system via automation
Automatically created calendar events for relevant opportunities
Logged outcomes to review accuracy and iteratively improve prompts

Tool Stack

Email service integration
Large Language Model for classification, extraction, and decision-making
Automation platform for workflow orchestration - BUBBLELAB.AI
Calendar integration

Visuals Included

Raw event email → structured AI output
Calendar entry generated by the assistant
High-level workflow diagram:

Inbox
  ↓
Event Detection
  ↓
Information Extraction
  ↓
Calendar + Tags

IMPACT

What Changed

Before

20–40 emails per week required manual review
Events were tracked inconsistently or missed
Inbox created constant decision fatigue

After

Event emails are processed automatically
Relevant opportunities appear directly on my calendar
Inbox shifted from task manager to passive notification stream

Measurable Outcomes (initial estimates)

⏱️ Time saved: ~2–3 hours per week previously spent scanning and organizing emails
📅 Event visibility: 100% of relevant events centralized in one calendar
📥 Inbox efficiency: Manual decision-making reduced for ~80% of event-related emails
🧠 Cognitive load: Fewer context switches and missed opportunities

Metrics will be refined as usage data accumulates.

Why This Project Matters

This project demonstrates:

End-to-end AI system design
Practical LLM usage beyond chat interfaces
Structured reasoning and decision-making with AI
Automation that produces real, measurable impact
A clear path from assistant-based systems to agentic architectures

Future Improvements

Split reasoning stages into independent agents
Add confidence scoring and fallback logic
Introduce feedback-based learning from calendar interactions
Expand relevance logic using historical preferences

Resume Version

`Built a modular AI assistant that processes learning-related emails every week, extracts structured event data using LLMs, evaluates relevance based on availability and topic alignment, and automatically creates calendar entries—saving ~2–3 hours per week and reducing inbox decision fatigue and allows me to not miss important events I would want to attend.`

AI Email Assistant for Events