Alumni AI Copilot & LinkedIn Intelligence Agent

AI Product Engineer · 2026 · 2 months · 1 person · 2 min read

Designed an AI research copilot using Exa AI and structured data enrichment to query alumni career trajectories for college Placement Officers (TPOs).

Overview

An AI intelligence platform built for Training & Placement Officers (TPOs). It indexes and enriches alumni databases using Exa AI web research agents, replacing manual LinkedIn searching with natural language queries like 'Find alumni with 5+ years experience at Microsoft'.

Problem

TPOs have unstructured networks of thousands of alumni, but manually finding relevant contacts for student placements or company relationships requires tedious multi-hour LinkedIn searches.

Constraints

  • Handle incomplete public LinkedIn profiles without breaking search precision
  • Corroborate research facts across GitHub, web evidence, and career snapshots
  • Prevent hallucinated career changes by implementing evidence verification

Approach

Integrated Exa AI web research tools into an autonomous agent pipeline. Designed a multi-step research loop: Profile Target $ ightarrow$ Exa Web Investigation $ ightarrow$ Structured Fact Extraction $ ightarrow$ Database Snapshot Comparison.

Key Decisions

Implement Shadow Mode / Research-Only Verification before automated actions

Reasoning:

Prevents automated agent hallucinations from publishing unverified career updates or outreach messages.

Alternatives considered:
  • Full autonomous execution with direct buffer publishing

Use Exa AI for structured web research over standard SERP APIs

Reasoning:

Exa provides neural semantic search and content extraction tuned for people and company intelligence.

Tech Stack

  • TypeScript
  • Python
  • Exa AI
  • LangGraph
  • Next.js
  • PostgreSQL
  • Prisma
  • OpenAI API

Result & Impact

  • Instant Natural Language
    Query Speed
  • Exa Neural Search
    Research Engine
  • Research $ ightarrow$ Corroborate
    Agent Pattern

Converted manual spreadsheet-based alumni tracking into an AI-powered natural language intelligence database.

Learnings

  • Validating simple useful loops first is vital — over-engineering complex agent memory before proving core retrieval leads to brittle systems.
  • An AI agent stating a fact doesn't make it true — evidence corroboration is mandatory.

AI + Web Research Architecture

Instead of simple keyword search, Alumni AI Copilot uses Exa AI to perform structured web research across public web sources, LinkedIn profiles, and GitHub activity.

Intelligence Pipeline

  1. Natural Language Query: TPOs ask complex questions (“Which alumni can help establish a hiring relationship with SAP?”).
  2. Exa Research Agent: Performs neural web searches, fetching career history, company changes, and public activity.
  3. Fact Corroboration: Compares raw web research against historical database snapshots to confirm recent job transitions.
All projects