ASK-ME

Inspiration Hiring managers lose hours every day to admin work — writing the same job descriptions, copying candidate data between systems, chasing calendar availability, sending templated emails. We built ASK-ME because the best recruiters shouldn't be spending their time on tasks a machine can handle. The insight was simple: people already live in Slack. Why make them switch to another tool?

What it does

ASK-ME is an autonomous AI recruiting agent that manages the full hiring pipeline through natural conversation in Slack. A hiring manager sends one message and the agent takes real action:

Writes complete job descriptions and posts them to Zoho Recruit or Workable
Pulls candidates from the ATS, scores them 0–100 against job requirements, and pushes results back
Checks Google Calendar availability, creates Google Meet links, and sends interview invites via Gmail
Reads resumes from Google Drive, creates Google Forms for job applications, and reads submissions
Remembers context across conversations — open roles, top candidates, preferences

Everything is accessible from a recruiter dashboard (Next.js, deployed on Vercel) that pulls live data from S3.

How we built it

The core agent is built with the Strands Agents SDK and deployed on AWS Bedrock AgentCore running Amazon Nova Pro. It exposes 31 tools across four categories: Core HR, Google (Gmail/Drive/Calendar/Forms/Sheets), and Zoho Recruit.

The Slack integration uses a two-Lambda architecture: a receiver Lambda verifies signatures and returns 200 within Slack's 3-second window, then async-invokes a worker Lambda that calls AgentCore and posts the response back. A third OAuth Lambda handles one-click Slack app installation — any workspace can add the bot in seconds via a hosted install page.

AgentCore Memory (Semantic + Summarization strategies) gives the agent persistent context across sessions. Tokens, credentials, and ATS data are stored securely in AWS SSM and S3. Infrastructure is managed via AWS CDK through the AgentCore CLI.

Challenges we ran into

Slack's 3-second timeout — solved with the async receiver/worker Lambda pattern
Google OAuth — required a full OAuth2 flow with refresh tokens; scope management across 5 APIs was non-trivial
ATS API quirks — Zoho Recruit requires a Client_Name field on job creation that isn't documented prominently; Workable's candidate update API uses a different auth model than their read API
AgentCore response parsing — the raw response format required recursive extraction logic to handle nested content structures reliably
Cold starts — solved by warming the worker Lambda every 5 minutes via EventBridge

Accomplishments that we're proud of

31 fully working tools deployed and tested end-to-end through Slack
True one-click install — any Slack workspace can add ASK-ME in under 60 seconds
The agent handles multi-step workflows in a single message: "pull candidates from Zoho Recruit, screen them, and schedule interviews with the top 3" actually works
AgentCore Memory means the agent genuinely remembers context — it knows your open roles without being told every time

What we learned

The Strands Agents SDK makes tool registration and agent orchestration remarkably clean — defining a tool is as simple as a Python function with a docstring
AgentCore handles the heavy lifting of runtime management, memory, and scaling in a way that lets you focus on the agent logic itself
Async Lambda patterns are the right architecture for any Slack bot that does non-trivial work
The hardest part of building an agent isn't the AI — it's the integrations

What's next for ASK-ME

Multi-user support with per-recruiter memory and permissions
Proactive notifications — "You have 3 new candidates for the Python Engineer role, want me to screen them?"
LinkedIn and Indeed integration for job posting distribution
Offer letter generation and e-signature workflow
Analytics: time-to-hire, pipeline conversion rates, sourcing quality