Portfolio
A complete overview of my projects, from foundational learning exercises to production AI systems.
Impact estimates are illustrative, modeled on the workflow each project targets, not measured client results.
SupportFlow Mini
A customer support ticket router with AI recommendation, human review gate, and webhook delivery. One problem, one model call, one human decision, one safe integration. Built with FastAPI, Pydantic schemas, SQLite, OpenAI, mock-first workflow design, and deterministic evaluation.
Est. impact: Complete workflow from customer ticket through AI recommendation, human review, to system integration with built-in metrics and evaluation.
RelayOps
A production incident workflow platform with grounded AI diagnosis. Ingests financial integration incidents, runs them through eight quality gates (environment, vertical slice, simulator, integrations, diagnosis, safe actions, evaluation, operations), and deploys with hardened security, observability, and recovery drills. Built end-to-end from empty folder to portfolio demo.
Est. impact: Complete incident workflow from detection through verified recovery, with deterministic AI that explains every claim.
Salesforce Intelligence
A retrieval-augmented chat assistant built on live Salesforce data, letting support and sales teams ask natural-language questions about accounts, cases, and escalations and get sourced answers instantly.
Est. impact: Turns a multi-record manual lookup across Cases, Accounts, and Opportunities into one sourced answer in seconds.
Deskline
An inbound phone menu system built with Python FastAPI and Telnyx Voice API. Routes customers through DTMF keypad prompts, plays business information via text-to-speech, and collects menu decisions. Handles state across multiple webhook events with SQLite persistence, webhook signature verification, and deterministic call testing before going live.
Est. impact: Complete call workflow from incoming call to menu response through call completion, fully tested without spending call credit.
Foundations
Projects I've built following core principles that power AI applications.
Ticket Triage Lab
A support ticket classifier you build from scratch. Start with rule-based keyword matching, then integrate OpenAI API, measure accuracy on labeled examples, and understand the boundary between your Python code and an external API. Ten lessons, eight to twelve study sessions.
Est. impact: Complete understanding of Python fundamentals, API integration, and evaluation metrics.
More projects
Narrower, faster builds shipped through Insolla, the same discovery-to-delivery process applied at a smaller scale.
Workflow Analyzer
Upload a process document, screenshot, or SOP and this tool extracts the workflow, then tells you exactly what to automate, which tools to use, and how many hours it will save each week.
Est. impact: ~2 hours of manual process mapping condensed into a 5-minute upload.
List Enricher
Upload a spreadsheet, describe the columns you want added in plain English, and get back a clean file enriched with real, web-researched data for every row.
Est. impact: Turns a multi-hour, row-by-row research task into one batch run.
SOP Generator
Describe a process in plain English or upload existing notes, and this tool turns tribal knowledge into a polished Standard Operating Procedure complete with roles, prerequisites, and quality checks.
Est. impact: Cuts SOP drafting from a half-day of writing to under 10 minutes.
Meeting Notes Analyzer
Paste in a transcript or raw notes and get a clean recap with action items, owners, due dates, decisions, and open questions ready to copy straight into your task tracker.
Est. impact: Turns 30 minutes of post-meeting note-cleanup into an instant recap.
Contract Analyzer
Upload any contract and get a plain-English summary, key dates, obligations on both sides, and severity-flagged risks worth pushing back on, all in about 30 seconds.
Est. impact: Cuts contract triage from ~45 minutes of manual review to under a minute.