Hi, I'm Ben Siegel.

ChoreStar

A gamified chore-tracking SaaS for families. Built solo from zero to 180+ web users and 2 paying customers, shipped across web and iOS, with the iOS app approved on its first App Store submission: market research, feature scoping, UX, full-stack development, Product Hunt launch, and growth. Behavioral incentives for kids, point systems, reward tracking, family dashboards. Shipped using LLM-assisted workflows with Claude, Codex, Cursor, and large language models.

I own the integration layer too: Stripe and PayPal webhooks handling payment and subscription events, Supabase real-time data sync, and a production AI feature on Claude with JSON-schema structured output, Zod validation, a rule-based fallback, per-user rate limiting, an eval suite that runs representative profiles against the live model, and a Playwright end-to-end suite covering the critical user flows. The AI feature runs on Claude Haiku 4.5. Paying customers on two continents, so pricing and onboarding have to make sense outside the US.

Hybrid Retrieval Pipeline & Eval Harness

A retrieval system over a 434-document personal corpus, roughly 836,000 words. Markdown-structure-first chunking that splits on heading boundaries and keeps the heading path as metadata, so every result cites its section: 10,356 chunks at a mean of 93 tokens. Retrieval is BM25 and cosine similarity fused with Reciprocal Rank Fusion, which fuses on rank and not on score, so the two retrievers never need a shared scale. Zero API cost, no network, and a deterministic embedder so an index rebuilds byte for byte.

The point of it was the evaluation. Twenty hand-verified cases split into verbatim and paraphrase families, reporting recall@k, MRR and precision per mode, so fusion has to beat each single retriever to earn its place. Three results contradicted my assumptions: plain BM25 beat the hybrid on recall, MRR and precision; paraphrase retrieval topped out at 0.141 MRR, which is the honest ceiling of a non-semantic embedder and the bar a real embedding API has to clear; and pseudo-relevance feedback made everything worse, so it ships defaulted off with the numbers in the comment and the code kept. The eval harness exits with a skip line when no index exists, because a green build on a test that never ran is worse than no test.

Partner Integration Portfolio

15 partner integrations connecting Practifi to the wealth-management ecosystem: Addepar, Schwab, Orion, Black Diamond, Tamarac, eMoney, Fidelity. Four of them built and maintained in-house. I own the lifecycle from intake through release sign-off: aligning partner release timelines against client migration windows, tailoring partner-built Salesforce apps to our data model, and maintaining the API and configuration documentation partners and clients build against.

I'm also the single voice for integrations in company-wide release forums, accountable for release readiness and the roadmap tradeoffs that go to the futures committee. I lead 2 engineers end to end on the roadmap.

Microsoft 365 Integration

An in-house integration on the Microsoft Graph API that delivers the client-facing email and calendar sync advisors use every day, currently being extended to SharePoint document sync.

Owning it in-house instead of consuming a vendor connector means owning the failure modes vendors hide: token expiry, throttling, partial sync, reconciliation after a failed batch. That's the part I'd want to talk about in an interview.

Growth Experiments & Client Onboarding

Ran an A/B experimentation program across growth experiments that lifted product engagement 25%, coordinating Engineering, Design, Marketing, and Sales through Agile to prototype, test, and launch. Instrumenting before shipping is cheaper than arguing about the results afterward.

Built a repeatable onboarding plan and template for new clients. Across 25–30 clients, typical engagements went from 4–6 weeks to 1–2, longer implementations from 8–12 weeks to 2–4, and the fastest recently down to a single day. How fast any one client goes comes down to the state of their data. Most of the delay was rework from unvalidated assumptions, not execution speed. Shipped production features in Apex and Lightning Web Components alongside it.

nCino & Financial Services Cloud Implementations

Built and configured Sales Cloud, Financial Services Cloud, nCino, and Experience Cloud portals for financial-services clients. Ran requirements gathering and current-state/future-state workshops, weighing feasibility, user experience, and business value in the solution design. Managed delivery across several concurrent engagements (roadmaps, sprint planning, release schedules) on Agile/Scrum teams.

This is the middle of the nCino arc. I built on it as a developer at U.S. Bank, configured it as a consultant here, and now own it as an ISV integration at Practifi. Very few Salesforce people have all three.

Customer Portal

Customer-facing engagement layer on Salesforce Community Cloud: loan status, document requests, the lifecycle work that turned a back-office tool into a self-service experience for borrowers.

Target.com Relaunch

First gig out of school, on a team of contractors rebuilding Target.com. Where I learned what production velocity looks like, and picked up the mobile campaign work that followed.