Portfolio
Practical systems, prototypes, and frameworks for clearer operations, bounded automation, and reliable AI-assisted workflows.
Project Index
Manufacturing Production System MES-style workflow architecture for production visibility, task sequencing, and operational control Type Project Focus Production visibility, ownership, sequencing, and handoffs Result / Output Relational production system with generated tasks and dependency controls
Business Problem
A small manufacturing workshop relied on paper notes, printed order sheets, and manual shop-floor tracking. As order volume increased, the team lost visibility into order status, production stage, task ownership, and sequential handoffs.
What I Built
Designed and implemented a relational Airtable production system connecting orders, order lines, production batches, production tasks, and production blueprints into one operational workflow.
System Logic
Blueprint-driven production logic, automated task generation, sequential dependency enforcement, and progress tracking across batch, order line, and order levels.
Tools
Airtable, relational database design, Airtable Automations, operator interfaces, workflow architecture.
Outcome
Created a structured production view where managers can monitor order progress and floor operators can follow a dependency-enforced task queue without manual status chasing.
Onsite Scheduling Architecture Reconstructed from a real-world scheduling problem using dummy data and generalized business rules. Type anonymized rebuild Focus Scheduling decisions based on incomplete assessment data Result / Output Validated estimates, team bookings, and auditable decisions
Business Problem
Hourly onsite service booking relied on judgment-based time estimates, inconsistent field-assessment inputs, and limited operating data, making reliable calendar automation premature.
What I Built
Built a sandbox scheduler that turns structured assessment data into validated duration estimates, team-specific bookings, and auditable outcomes.
System Logic
Normalize intake data → validate readiness → estimate duration → place the job in an available team calendar → record the decision.
Tools
Python, Pydantic, SQLite, Flask, FastAPI, pytest, Sentry.
Outcome
Demonstrated an explainable scheduling architecture that books ready jobs, skips incomplete jobs without guessing, and preserves an append-only audit trail.
Workflow Automation Readiness Framework Diagnostic framework for determining whether a specific workflow is ready for reliable automation. Type Framework Focus Whether a workflow has the conditions required for reliable automation Result / Output Scored assessment and dependency-ordered remediation guidance
What It Assesses
Whether a workflow has the process visibility, structured data, ownership, controls, measurement, and adoption conditions required for reliable automation.
Readiness Model
Five preconditions: single source of truth, structured reusable data, process measurement, clear ownership, and control gates.
Method
Representative-path modeling, triangulation across interviews, observed behaviour, and system structure, plus validation playback and root-cause analysis.
Output
A scored readiness assessment, dependency-ordered remediation sequence, clean-order metric, and guidance on where bounded automation can safely begin.
AI Readiness Framework Tiered framework for moving from unmanaged AI adoption to deliberate AI adoption. Type Framework Focus Companies act on AI before they can clearly see their processes, data, metrics, and existing AI exposure. Result / Output A tiered path from operational legibility to standardization, deterministic automation, and bounded AI-assisted operations.
What It Assesses
Whether a company is ready to adopt AI deliberately: process legibility, data trust, measurement, existing AI exposure, standardization, automation readiness, and where AI is actually justified.
Readiness Model
Four-tier model: Operational Legibility, Standardization, Deterministic Automation, and AI-Assisted Operations.
Method
AI-adjacent business diagnostic, process mapping, system-of-record review, metrics baseline, data hygiene review, and inventory of sanctioned and unsanctioned AI usage.
Output
A clear current-state diagnosis, tier placement, sequencing guidance, and a practical next step — including when the right next step is not AI.
Slack Bot for Real Estate — Buyer Preference Collection Working Slack MVP for structured buyer intake Type Prototype Focus Fragmented buyer conversations and incomplete preference records Result / Output Structured buyer intake and confirmation inside Slack
Business Problem
Real estate buyer intake is often handled through scattered conversations, repeated follow-up messages, and incomplete preference notes. This slows qualification, weakens handoffs, and makes it harder for agents to maintain consistent buyer profiles.
What I Built
Built a working Slack bot that collects buyer preferences through an interactive modal and sends a structured confirmation message after submission.
Core Functionality
The bot collects search area, price range, bedroom requirements, and must-have notes. It supports a /buyer-profile slash command, interactive modal submission, confirmation DM, hello command, and error handling.
Technical Implementation
Implemented with Node.js, Express.js, Slack Bolt / Web API, HTTP endpoints, environment configuration, ngrok webhook tunneling, and HMAC-based Slack request signature verification.
Outcome
Created a functional buyer intake workflow that turns informal buyer conversations into structured preference data inside Slack.
AI Lead Triage + Autoreply Workflow Zapier workflow for labeled Gmail intake, AI lead classification, Slack alerts, Google Sheets logging, and branch-specific response handling. Type Workflow automation Focus Manual lead prioritization and repetitive first-response drafting Result / Output Classified lead paths, structured logging, Slack alerts, and Gmail response handling
Business Problem
Inbound sales inquiries arrived with different levels of urgency, clarity, and fit. Without triage logic, each message had to be manually reviewed, prioritized, logged, and answered.
What I Built
Built a working Zapier automation that starts from a controlled Gmail label, preprocesses inbound email text, normalizes fields with JavaScript, and routes messages through either a bypass path or an LLM-based classification path.
Workflow Logic
Gmail labeled email trigger → text cleanup/truncation → JavaScript normalization and bypass check → LLM classification → hot/warm/cold path routing → Gmail response/draft handling → Slack notification → Google Sheets log.
Design Detail
The key design choice was separating deterministic routing from AI classification. The workflow first checked whether a message belonged in the intake flow before sending relevant messages into LLM-based classification.
Tools
Zapier, Gmail, Formatter by Zapier, Code by Zapier, JavaScript, OpenAI API, Slack, Google Sheets.
Outcome
Created a shippable AI-assisted lead triage workflow that standardized first-response handling, surfaced high-intent leads, and created a lightweight operational record of inbound lead activity.
Appointment Reminder + Review Flow Workflow automation for appointment follow-up and review collection Type delivered automation Focus Consistent appointment reminders and post-service review requests Result / Output Automated SMS reminders, email follow-up, and tracking
Business Problem
A service business needed a more consistent way to reduce missed appointments and collect customer reviews without manual follow-up.
What I Built
Built a functional automation workflow that sent SMS reminders before appointments and email review requests after service completion.
Workflow Logic
Appointment intake → scheduled SMS reminder → post-appointment email follow-up → review request → Google Sheets tracking.
Tools
Make, Zapier, Google Forms, Google Sheets, Telnyx, Gmail.
Outcome
Delivered a deployment-ready automation workflow and clarified the operational requirement for reliable automation: standardized appointment and calendar data.
RUDE.AI — Bounded LLM Reminder System Telegram-based reminder product with a bounded AI workflow Type Live AI product Focus Bounded LLM generation, safety-gated input handling, timezone-aware scheduling, billing webhook handling, privacy-conscious retention Result / Output Escalating reminders across a guarded 24-hour product lifecycle
What I Built
Built a live Telegram-based AI reminder product that turns one user task into a scheduled sequence of escalating reminders across a 24-hour lifecycle.
System Logic
The system keeps the LLM inside a narrow product loop: fail-closed input moderation before storage or generation, prompt-role separation, timezone-aware schedule computation, transactional billing webhook handling, and privacy-conscious data retention.
Tools
Python, FastAPI, python-telegram-bot, APScheduler, PostgreSQL, SQLAlchemy, Alembic, OpenAI API, Lemon Squeezy, Railway, pytest.