Selected transformations. Anonymized where it matters.

Cybersecurity · Software Supply-Chain Security

Product Build · Sales Enablement

A software supply-chain security company

• Python · FastAPI
• TypeScript · React
• PostgreSQL

A sales-demo mockup turned into a working multi-tenant product the team demos live.

The challenge

Vendors in supply-chain security have to prove value fast. Attacks on open-source components doubled in 2024 (Sonatype), and 81% of organizations report at least one supply-chain breach in the past year (BlueVoyant).

What I delivered

Their sales team was demoing from a static mockup. I converted it into a working multi-tenant product the team logs into with prospects: 12 functional views, auth and RBAC, live catalog ingestion, CVE/VEX security feeds, package repository endpoints, and a one-command demo bootstrap so every call starts from a clean, presentable environment.

Cybersecurity · Software Supply-Chain Security

Data & Intelligence Platform

A software supply-chain security company

• Python · FastAPI
• TypeScript · React
• PostgreSQL · Celery
• Claude

Six signals, one account health score: a Customer Success intelligence platform.

What I delivered

Their CS team was flying blind across six sources. I built a platform that syncs Salesforce, Gong, HubSpot, Pylon, ZoomInfo, and product usage signals into a four-dimension health score for every account, generates AI account briefs, and gives CS, AE, and SE teams role-based workspaces with alerting and scheduled sync workers.

Cybersecurity · Software Supply-Chain Security

Risk Framework · Research Engineering

A software supply-chain security company

• Python
• OpenSSF Scorecard
• OSV.dev · CISA KEV

An open-source risk framework that scores the 100 most-used Java projects from live data.

The challenge

Most open-source risk assessment is done by hand, from stale snapshots. This framework replaces that with live, confidence-adjusted scoring.

What I delivered

A framework that scores the 100 most-used Java open-source projects across 7 risk dimensions using live data from GitHub, OpenSSF Scorecard, deps.dev, OSV.dev, CISA KEV, and Maven Central, with confidence-adjusted scores and an interactive report that flags actively exploited CVEs.

Real Estate · Technology

AI Product Build

A real-estate intelligence platform

• TypeScript · Fastify
• React 19
• SwiftUI · iOS
• Claude

Find the best real estate opportunities for what you need, scored in real time and explained in plain language.

The challenge

Over 90% of leading real-estate firms now call AI a strategic priority (Colliers), yet buyers, investors, and agents still compare listings across sites and spreadsheets, one property at a time.

What I delivered

A realtor management platform that lets buyers, investors, and agents scan the market against their own needs. Each user sets their markets, budget, property criteria, and financial goals, and the platform surfaces the strongest opportunities, with saved filters and alerts when something worth acting on appears.

Every opportunity gets an AI analysis across dozens of factors in real time, covering financial performance, comparable sales, rent estimates with confidence levels, neighborhood quality, and local regulation. Each parameter receives its own score, calculated deterministically so the same inputs always return the same result, and AI explains the reasoning in plain language.

The results land in a clear dashboard, on the web and on iOS, that adapts to the person using it: an investor sees returns and risk, a homebuyer sees fit and affordability. It brings the scores, risk flags, and offer strategy together so they can compare options and make a decision, while agents manage multiple clients, showings, and pipeline in the same system.

Consumer Technology · Crowdfunding

RAG Knowledge Platform

A leading crowdfunding platform

• Python · FastAPI
• pgvector
• Slack Bolt
• OpenRouter

Product knowledge for every team, in Slack: current, cited, verified, with escalation SLAs.

The challenge

Product knowledge lives across Confluence, Drive, Jira, Zendesk, and Slack, and customer-facing teams lose time hunting for it. A self-service answer costs $1.84 versus $13.50 with an agent involved (Gartner), and 45% of support interactions include someone searching for knowledge mid-conversation (Verint). An assistant that answers from stale or unverified content just moves the problem.

What I delivered

A retrieval-augmented (RAG) product knowledge platform that teams use directly in Slack. It ingests Confluence, Google Drive, Jira, Zendesk, and Slack history, so answers are grounded in up-to-date source content rather than a model's memory. Every answer carries source citations, an authority tier, and a confidence score.

Answers are verified before they are sent. A 10-step pipeline combines hybrid retrieval, authority re-ranking across four source tiers, cross-encoder re-ranking, multi-model routing, and per-citation verification. Billing and pricing questions need two independent sources.

When the system can't verify an answer, it escalates under an SLA: the right product manager is identified from a topic-owner map, notified, and their expert reply is routed back to the asker. A full admin portal covers queries, escalations, and topic-owner routing.

Online Gaming

Product Build + Research Engineering

An online gaming operator

• Rust
• PWA · Service Worker
• Anthropic API

A game-theory training product for players, and the production-grade Rust engine behind it.

The challenge

Game-theory-optimal training has become table stakes in the poker ecosystem: an entire commercial category now competes on the quality of its math. The differentiator is owning the engine rather than licensing one.

What I delivered

The engine: a Counterfactual Regret Minimization solver for Texas Hold'em in production-grade Rust. Multi-street solving from pre-flop through the river, CFR/CFR+/DCFR variants, exact game-theoretic solutions with an independent best-response validator that catches shared bugs, and automatic exploitability reporting.

The product: an installable iPhone PWA training app on top, with mixed-strategy GTO analysis, multiple learning AI opponents that adapt as they train players, Hold'em and Omaha support, offline play, and an AI coach. No App Store, no review cycle.

Media · Research

Platform Build · Trust & Safety

An online news-research institute

• TypeScript · Fastify
• Next.js
• OpenRouter
• pnpm monorepo

Claims in, verdicts out, every decision auditable: a claim-verification platform.

The challenge

At one major fact-checking organization, the share of checked claims involving AI-generated content jumped from 7% to 16% in a single year (Reuters Institute). Verification teams need pipelines, not heroics.

What I delivered

An end-to-end claims platform: public claim submission flows through an intake gateway into a review queue, a deterministic Trust Engine produces initial assessments, and human reviewers work an admin verdict browser. Multi-model LLM routing supports the analysis with no single-vendor dependency. Built as a typed monorepo with generated API contracts shared between services, and governance documented up front: a legal and liability framework, a threat model, and 13 architecture decision records.