A 90-minute build by Anuj Kapoor

How does the CEO of an $8B company triage 120 LinkedIn comments without losing an afternoon?

A hiring post pulled in 120 comments asking to apply. Reading each profile, scoring each comment against the role's criteria, is a 2-6 hour job. I built an agent to do the first pass. Ranked the field in 4 minutes.

Commenters analyzed
119
Worth interviewing first
9 (8%)
Strong fit
6
Top score
62/100

The scoring rubric

Five axes derived from the post itself, weighted by what the role calls for:

These weights and definitions are tunable. Different criteria → different rankings. The agent runs the same way against any rubric you define.

What the commenter pool actually looks like

Top role keywords (qualified)

ai
16
engineer
8
product
8
agent
7
ml
3
founder
3
ops
2
automation
2

Top geographies

India
4
Australia
3
Singapore
3
United States
2
Hong Kong SAR
2
New Zealand
1
New York City Metropolitan Area
1
Nigeria
1

Top 10 candidates - anonymized

Identities, profile links, and full breakdowns sit in the private brief. Below shows score, tier, and the agent's reasoning so you can see how it's evaluating.

#ScoreTierWhy qualified
162strong fit14-year full-stack engineer with hands-on LangGraph/LangChain multi-agent RAG systems, CKA, and production SaaS architecture experience.
262strong fitProduction RAG pipeline builder with strong cloud infra/DevSecOps background, compliance experience in fintech, and a concrete technical article shared.
358strong fitBuilt a multi-agent system with LLM-as-judge, conditional retries, MCP tool layer - demonstrates real agentic architecture thinking.
458strong fitProduction ML background at a bank, Google Build With AI Hackathon winner, and co-founded an LLM/agent startup on the back of it.
554strong fitClaims hands-on multi-agent orchestration platform with named agents, routing logic, memory, and self-correction running on owned hardware.
652strong fit11+ years AI experience, architecting production agentic RAG systems at Shell serving 15K+ daily users with LangChain/LangGraph/CrewAI stack.
744worth a lookHands-on with n8n, LangChain, CrewAI, MCP servers, and Python; built his own agent-readiness platform RankedLM.
842worth a lookSolid Technical Lead at AI-focused company (Harrison.ai) with multi-language backend depth and team architecture experience.
942worth a lookBuilt and shipped a conversational AI chatbot for agentic payments on WhatsApp and automated a $200M+ OTC treasury workflow - concrete deliverables.

How it was built

  1. Pull the public comments on the post + each commenter's public profile data
  2. Score each commenter against the 5-axis rubric using Claude Sonnet 4.6, structured JSON output per candidate
  3. Generate two briefs: named (private, sent on request) + anonymized (this page)
  4. Static site (Next.js + Tailwind), deployed on Vercel
Build time
~90 min
Total cost
~$2
Manual triage equivalent
2-6 hrs

Build included one tool pivot mid-way. Net: a CEO's 2-6 hours of focused review work compresses to ~$2 and 4 minutes of agent runtime, with the human still making the actual hiring decision.

The named ranking

The full named brief - top 25 candidates with profile URLs, full comment text, and per-axis breakdowns - is available on request. Happy to forward it over LinkedIn if it's useful.

Want to build this for your own team?

The same agent can score any LinkedIn post's commenters against any rubric you define - hiring, partnership outreach, event audience analysis, customer pre-screening.

Drop a comment on the LinkedIn post and I'll share the build walkthrough.