Senior Ruby on Rails & AI Product Engineer - I build and run AI products end to end.

What it is

A two-sided marketplace where anyone can post a CV or a job, for any role, in any country. Employers search candidates with one natural-language box and pay only to unlock contact details. On top sits a recruitment CRM for agencies: a private candidate book, bulk import, pipelines, document tracking, and client shortlists. Built from an empty repo to production in about two months.

Key contributions & impact

• Semantic search: Natural-language candidate and job search on OpenAI embeddings in pgvector with HNSW indexes, a similarity floor, cached query vectors, and trigram-indexed keyword paths.
• LLM CV parsing: Reads PDF, DOCX, and image CVs into a structured profile with structured outputs, a retry ladder, stall detection, a parse cache, and per-call cost tracking. It also turns photographed job posters into draft postings.
• WhatsApp AI assistant: Built directly on Meta's Cloud API: a signature-verified webhook with an idempotency ledger, passwordless sign-in, reverse-OTP verification, and a tool-calling LLM assistant for CV intake, job posting, and candidate search.
• Multi-tenant CRM: Isolated each agency's candidate book with one tenancy column and scopes instead of a re-architecture. Bulk CV/ZIP/spreadsheet import with dedupe, pipelines with an append-only audit trail, and document-expiry tracking.
• Metered billing: Per-company contact credits with a consent-based access model, row-locked spending that can't overspend under concurrency, UPI reconciliation, and tiered plans with trials.
• Production: Two Hetzner servers on a private network behind Cloudflare, deployed with Kamal 2 and GitHub Actions, monitored by Sentry with no personal data sent, and backed by ~6,500 automated tests.

What it is

A production AI system that watches 150+ official sources (sanctions authorities, central banks, securities and trade regulators) and decides which items are executed, binding changes. Each one becomes a structured decision brief: what changed, why it matters, who must act, and which deadlines to watch.

Key contributions & impact

• Multi-stage LLM pipeline: A regex pre-filter, four-gate LLM triage, and JSON-schema-validated summarisation with revision loops, versioned prompt caching, and tiered model selection to match cost to each task.
• Ingestion at breadth: 40 custom regulator scrapers (OFAC, FinCEN, MAS, HKMA, RBI, SAMA, CBUAE, UN Security Council and more) alongside tiered RSS ingestion, PDF extraction, and deduplication.
• Search and personalisation: Embedding-based semantic search, a "Read next" recommender that suppresses duplicate coverage, and user-vector personalisation.
• Performance: Cut semantic search latency 18× (1,540 ms → 87 ms) through profiling-led caching of the vector matrix.
• Clarify with AI: Grounded, streaming Q&A on every brief, with PII masking and tiered monthly quotas tied to billing and trials.
• Full ownership: Backend, infrastructure, CI/CD, email delivery, SEO and AI-answer-engine visibility, and analytics that separate human visits from crawler traffic.

What it is

A production backend powering large-scale research ingestion, semantic search, and retrieval workflows with a strong focus on performance and reliability.

Key contributions & impact

• High-performance services: Designed backend services handling large datasets with consistently low response times.
• Semantic retrieval: Built vector database pipelines to support semantic search and retrieval use cases at scale.
• Infra + delivery: Owned infrastructure provisioning, deployments, and delivery using Terraform, Ansible, and automated CI/CD pipelines.

What it is

A cross-platform mobile application supporting real-time collaboration and offline-first research workflows.

Key contributions & impact

• Single codebase: Built a shared iOS/Android codebase with clean, maintainable API integration.
• Offline-first reliability: Implemented robust sync patterns to ensure usability under poor network conditions.
• Performance tuning: Optimized performance and data handling for field usage and low-connectivity scenarios.

What it is

A professional networking platform designed for performance, scalability, and long-term operational stability.

Key contributions & impact

• Reliability: Maintained 99.99% uptime using zero-downtime deployment practices.
• Scaling: Scaled to thousands of active users through caching and background processing.
• Product capabilities: Built secure content delivery and real-time notification systems.