AI / APPLICATION ENGINEERING

AI-Powered Analytics Assistant

An interactive AI portfolio assistant combining generative AI, curated professional context, job-description analysis, and Google Calendar scheduling within a production-style Next.js application.

OpenAI APINext.jsTypeScriptPrompt EngineeringContext DesignGoogle Calendar APIGoogle OAuth 2.0Google MeetVercel

Business Challenge

Portfolio visitors and recruiters often need to understand a candidate's experience quickly, compare skills against a job description, identify relevant projects, or determine whether a conversation would be useful.

A static website can display this information, but users still have to search through multiple sections manually. The goal was to create an interactive assistant that can interpret questions, use verified professional context, generate useful answers, and perform meaningful actions such as scheduling a meeting.

Visitors may not know where relevant experience is located
Recruiters need faster role-to-profile comparison
Static portfolio pages cannot answer follow-up questions
Unstructured profile information can be difficult to summarize
Scheduling a conversation typically requires additional back-and-forth
AI responses must remain grounded in verified experience
CURRENT CAPABILITIES

What Abhilash AI Can Do

The assistant combines conversational AI with application workflows. It does more than generate text by routing user intent into different experiences.

💬

Profile Q&A

Answers questions about experience, skills, certifications, projects, technologies, and career background using curated profile context.

🎯

Job Match Analysis

Compares a job description against profile evidence and identifies strong matches, relevant experience, and realistic gaps.

📅

Meeting Scheduling

Collects meeting details, checks Google Calendar availability, presents valid time slots, and creates a confirmed Google Meet event.

APPLICATION ARCHITECTURE

AI Assistant Architecture

The application separates the user interface, server-side API routes, verified professional context, AI processing, and external actions such as Google Calendar scheduling.

STEP 01
👤

Visitor

Natural-language question or action request

STEP 02

Next.js Application

Chat interface and server-side API routes

STEP 03
🧠

Context & Intent

Profile knowledge, prompts, and scheduling state

STEP 04
AI

OpenAI

Language understanding and structured response generation

STEP 05

Action Layer

Chat response, job analysis, or calendar workflow

Core Conversation Flow

User Question
Next.js API Route
Profile Context
OpenAI
Structured Response

How the Assistant Works

01

User Intent

The visitor asks a question or selects an action such as analyzing a job description or scheduling a meeting.

02

Context Preparation

Verified portfolio, experience, skills, certifications, and project information are supplied to the model.

03

Prompt Construction

System instructions define the assistant's role, permitted claims, response style, and workflow rules.

04

Intent-Specific Processing

The application routes the interaction into conversational Q&A, job analysis, or scheduling logic.

05

External API Action

When scheduling is requested, server-side APIs interact with Google Calendar rather than relying on the language model to invent availability.

06

Confirmed Response

The application displays the generated answer or confirms an external action only after the backend reports success.

Portfolio Assistant Use Cases

The current portfolio implementation supports multiple recruiter and visitor interactions from the same conversational interface.

EXPERIENCE Q&A

“What experience does Abhilash have with Microsoft Fabric?”

The assistant identifies relevant verified experience and summarizes Fabric work around Lakehouses, pipelines, semantic models, migration initiatives, and analytics.

JOB MATCH

“How well does this job description match Abhilash?”

The assistant compares job requirements with supplied profile evidence and provides strong matches, relevant experience, potential gaps, and a realistic overall assessment.

SCHEDULING

“I'd like to schedule a meeting with Abhilash.”

The assistant starts a guided scheduling flow, checks actual calendar availability, presents valid meeting times, and creates the event only after explicit visitor confirmation.

AI Job-Matching Workflow

One of the assistant's most useful recruiter-facing capabilities is comparing a job description against verified candidate information.

Job RequirementProfile EvidenceAssessment
Microsoft FabricLakehouse, pipelines, semantic modelsStrong Match
SQLData engineering and analytics workloadsStrong Match
Power BIDashboards, semantic models, KPI reportingStrong Match
Unlisted TechnologyNo supporting profile evidencePotential Gap

Example assessment shown for portfolio demonstration.

GOOGLE CALENDAR INTEGRATION

AI-Assisted Meeting Scheduling

Scheduling is handled as an application workflow rather than a fabricated AI response. The chatbot gathers meeting information, calls server-side scheduling APIs, checks the connected Google Calendar, and creates a real event only after validation and user confirmation.

Scheduling Architecture

Visitor
ChatBot
Availability API
Google Calendar Free/Busy
Available Slots
User Confirmation
Schedule API
Calendar Event
Google Meet
01

Collect Meeting Details

The assistant collects the visitor's name, email, optional company or reason, preferred date, and scheduling context.

02

Interpret Date

Natural-language inputs such as tomorrow, Friday, or next Monday are converted into a specific calendar date.

03

Confirm Date

The visitor confirms the interpreted date before the application performs an availability check.

04

Check Availability

A server-side availability API queries Google Calendar Free/Busy data and generates valid 30-minute meeting slots.

05

Select Time

Only available business-hour slots are presented to the visitor for selection.

06

Revalidate & Book

Availability is checked again immediately before booking to reduce double-booking risk.

07

Create Calendar Event

The scheduling API creates the meeting in Google Calendar and adds the visitor as an attendee.

08

Generate Google Meet

The confirmed event includes a Google Meet conference link and calendar invitation.

Scheduling Controls

Natural-language date parsing
Explicit date confirmation
Google Calendar Free/Busy validation
30-minute meeting duration
Weekday and business-hour validation
Mountain Time scheduling logic
Past-date and past-time protection
Availability rechecked before booking
Double-booking conflict handling
Google Meet conference generation
Calendar attendee invitation
Success shown only after backend confirmation

Prompt, Context & Guardrail Design

The quality of the assistant depends on both the information supplied to the model and explicit instructions controlling what it may claim or do.

System Instructions

Define the assistant's role, response style, workflow behavior, and boundaries around unsupported claims.

Structured Context

Provide verified experience, skills, certifications, projects, technologies, and professional-impact statements.

Action Guardrails

The assistant does not claim that a meeting was scheduled, email was sent, or external action succeeded until the corresponding backend API confirms the action.

Current Architecture vs Future RAG

The current portfolio implementation uses curated profile and resume-related context supplied directly to the model. It does not represent the current implementation as a retrieval-augmented generation system.

CURRENT IMPLEMENTATION

Context-Grounded AI

✓ Curated profile context
✓ Resume and project knowledge
✓ Prompt engineering
✓ OpenAI API
✓ Job-description comparison
✓ Google Calendar scheduling
✓ Google Meet generation
FUTURE ENTERPRISE ARCHITECTURE

Retrieval-Augmented AI

○ Document retrieval
○ Vector-based semantic search
○ RAG architecture
○ Enterprise data connectors
○ Governed semantic-model access
○ Role-based authorization
○ Evaluation and monitoring framework

Security & Responsible AI Design

AI and external API integrations are kept server-side so application credentials do not need to be exposed to the browser.

OpenAI API key stored server-side through environment variables
Google OAuth credentials stored outside browser-side code
Browser communicates with protected Next.js API routes
Calendar availability is queried server-side
Assistant is instructed to use supplied professional context
Unsupported experience should not be invented
Calendar events are created only after explicit confirmation
Availability is revalidated before meeting creation
Private calendar event contents are not exposed to visitors
Production deployment can add rate limiting and bot protection

Technology Stack

Application

Next.js

React

TypeScript

Tailwind CSS

AI

OpenAI API

LLM

Prompt Engineering

Context Design

Integrations

Google Calendar API

Google OAuth 2.0

Free/Busy API

Google Meet

Backend & Deployment

Next.js API Routes

Environment Variables

GitHub

Vercel

Business & Portfolio Value

Provides an interactive interface to professional experience and projects
Helps recruiters find relevant experience without manually scanning every page
Supports realistic comparison between job requirements and candidate evidence
Demonstrates practical generative AI integration rather than a static chatbot demo
Connects AI conversation with a real external workflow through Google Calendar
Reduces scheduling back-and-forth by presenting actual available meeting times
Demonstrates API integration, OAuth, server-side workflow design, and frontend UX
Creates a reusable foundation for more advanced enterprise AI assistants

Future Enhancements

The current assistant is intentionally grounded in curated context. A future enterprise implementation could extend the architecture with retrieval, governed business data, stronger authentication, and production observability.

Resume and document retrieval
Vector-based semantic search
RAG architecture
Power BI and semantic-model integration
Enterprise authentication
Conversation analytics
Usage and cost monitoring
Model evaluation framework
Role-based data access
Rate limiting and bot protection
Scheduling horizon controls
Production monitoring and alerting
CASE STUDY TAKEAWAY

AI becomes more useful when it can understand, reason, and act.

The project goes beyond adding a chatbot to a portfolio. It combines verified context, generative AI, job-requirement analysis, structured workflow logic, and a real Google Calendar integration to create a more useful and interactive application.

This case study represents the current portfolio assistant implementation and related application patterns. Retrieval-augmented generation, vector search, enterprise business-data connectors, and governed semantic-model access are presented as future enhancements and are not represented as currently implemented production features.