Projects

Problem

Researching an account meant switching between LinkedIn, data tools and the prospect's own website. Multi-step email and LinkedIn sequences were tracked by hand, so steps were missed or sent late; high-intent website visitors stayed anonymous; and leads, calls, audits and outreach status lived in separate places.

What I built

One web app that discovers companies worth pitching, qualifies them with AI (website audit, ICP fit, keyword audit, battle card), enriches them with firmographic and LinkedIn data, and engages them through a scheduled 4-step LinkedIn + email sequence with human approval of every draft. An autonomous agent finds new ICP-fit companies, inbound visitors are de-anonymized with an AI intent summary, and everything lands in a lightweight CRM with one-click weekly reports.

Key features

• Outreach engine: 4-step LinkedIn + email sequence with human-approved drafts
• Per-prospect AI audit (Content / AEO / GEO), ICP fit and battle card
• Inbound intelligence: de-anonymized visitors with AI intent summaries
• Autonomous lead-discovery agent with pitch angles
• Domain research, CRM call tracking and weekly reports

Engineering

A PostgreSQL advisory lock keeps the scheduler single-owner; each draft is claimed with a conditional UPDATE so concurrent workers never double-send; overdue steps are skipped instead of sent as a backlog blast; every user-supplied URL is validated hop by hop against public IP space (SSRF protection); Google OAuth restricted to the company Workspace, role-based access and an audit log; a REST API and an MCP server with hashed, revocable API keys.

Problem

Hiring lived in spreadsheets, email threads and chat, so nobody could see where a candidate stood. Hundreds of resumes per role were read one by one, new roles opened with no record of who approved the budget, and offer letters were edited by hand in Word.

What I built

An internal platform that runs the full journey: a hiring-request workflow (request → HR review → budget approval → posting), AI scoring of every applicant against the job's requirements with a ranked top-20 shortlist and an explanation for each ranking, interview rounds on a pipeline board, a document generator with 9 PDF templates, and notifications routed by role to in-app, email or a personal Slack DM.

Key features

• AI applicant scoring and a ranked top-20 shortlist with per-candidate reasoning
• Hiring-request workflow with stage tracker, approvals and audit trail
• Interview rounds and bulk stage moves on a pipeline board
• Offer letters and contracts generated to PDF with live preview
• Notifications routed by role: in-app, email or Slack DM

Engineering

Chose the scoring model by benchmarking 8 language models on real evaluations. Personal data is encrypted at rest (Fernet); notification delivery is queued, retried and de-duplicated; PostgreSQL with hand-written Alembic migrations.

Problem

Work for several clients was spread across content, SEO, design and development teams, with no single place to see who was doing what, what was waiting on verification or approval, and how each team was performing.

What I built

A multi-role operations platform: clients and their content, team spaces for writers, SEO, design and development, a task workflow with acceptance, verification, approval and rework stages, workload and team-performance views over any date range, reports, a feedback inbox with analysis, an audit log and a superadmin console.

Key features

• Role-based dashboards for superadmins and team members
• Task workflow: assigned → in progress → verification → approval, with rework
• Team workload and performance over 7, 30 or 90 days
• Client spaces with Google Analytics / Search Console integrations
• Feedback inbox and analysis, reports and an audit log