HelpMeRad
AI-powered radiology report generation that learns from each radiologist's style. Few-shot learning from gold-tier examples, multi-layer directive enforcement, AI failover, full generation audit trails — built for a domain where clinical precision is not optional.
The challenge
Radiologists spend hours dictating structured reports. Each modality has different conventions, each radiologist has their own style, and each institution has its own compliance requirements. The client needed an AI system that matches a specific radiologist's voice while enforcing clinical directives — built HIPAA-compliant from day one.
What we built
- Few-shot learning from gold/silver/bronze quality-tiered examples with matchScore ranking
- Multi-layer directive enforcement — prompt injection, post-generation compliance scoring, annotated example reasoning
- 7+ exam categories (MRI, CT, Ultrasound, DEXA, Elastography, Renal Doppler, MRI Prostate)
- AI failover — Claude primary, automatic xAI Grok fallback with provider transparency
- Full generation audit trail — prompt, examples, directives, compliance JSON per report
- Post-generation pipeline — corrections, deterministic rules, medical reconciliation
- HIPAA architecture + three-environment CI/CD
Stack
Phased delivery
Phase 1 — MVP COMPLETE Train-by-example workflow Template CRUD + report generation Multi-modality (CT, MRI, US, DEXA) Production deploy + HIPAA architecture Nov 2025 — Dec 2025 (6 weeks) Phase 2 — Rad AI Assistant COMPLETE Few-shot learning from ranked examples Target report quality tiers (gold/silver/bronze) AI failover (Claude → Grok) HIPAA compliance review pass v2.0.0 stable production release Dec 2025 — Jan 2026 (4 weeks) Phase 3 — Precision & Enforcement ACTIVE Multi-layer directive enforcement engine Per-generation compliance audit trail Style guide versioning + refactoring Auto-detect exam type from input Post-generation rules + corrections pipeline Compliance transparency UI (violations) Feb 2026 — ongoing
What governance looked like
Every phase tracked via work items with acceptance criteria. Production deploys held until staging verification passed. The directive engine alone had 11 acceptance criteria across three layers — each verified against audit logs before shipping.
“I had the concept and I’d been trying to build it myself with Claude — I could get it to generate reports in a chat window, but I hit a wall turning that into something real. Travis and The Collective took that idea and built a product I use every day to do my actual work. Now I’m ready to put it in front of other radiologists.”— Adam Brochert, MD · Radiologist & HelpMeRad Founder
What AI made possible
The AI learns from each radiologist's own curated reports — not generic fine-tuning, but few-shot learning ranked by quality tier. A three-layer directive engine ensures clinical rules are followed, and a transparency banner shows exactly what was applied. The difference between "AI that writes reports" and "AI that writes reports a radiologist would actually sign."
HealCircle
A wellness practitioner marketplace with an AI Healing Guide, a karma-based economy, and safety guardrails that escalate to 911 — built on a timeline dictated by a real-world event: HealFest.
The challenge
A wellness marketplace connecting practitioners to consumers — launched in five weeks for a real-world event (HealFest). The hardest problem: an AI guide that's helpful without being medical, recommends without diagnosing, and escalates genuine emergencies to 911.
What we built
- Practitioner directory with search, filter, and profile pages
- Practitioner profile CRUD — modalities, insurance, pricing, free resources
- Clerk authentication — email, phone, magic links (free tier)
- Branded responsive design (dark green gradient + coral orange)
- AI Healing Guide with two-path entry: "Relieve Symptoms" and "Prevent and Live Longer"
- Three-tier recommendation engine: Immediate Relief, Short-Term Plan, Rule-Out / Refer
- Safety guardrails: red flag detection, 911 escalation for genuine emergencies
- Ephemeral conversations — no server-side PHI storage by design
- Karma Economy: double-entry ledger, points, vendor offers, referral bonuses
- Stripe payment integration — vendor items, checkout, webhooks
- Admin dashboard: vendor activity, karma sessions, active offers
- Full blog CMS with WordPress migration — 686 posts, categories, tags, SEO metadata
- Consumer review system — testimonials with privacy controls (full name / first name / anonymous)
- Practitioner application + approval workflow with access code generation
- Promo code system with CSV import, usage tracking, and test codes
Stack
The AI Healing Guide
Personality: empathetic, Brené Brown-inspired Two paths: "Relieve Symptoms" (acute) "Prevent & Live Longer" (proactive) Recommendations (3-tier): 1. Immediate relief — what to try today 2. Short-term plan — practitioners to see 3. Rule-out / refer — when to see an MD Safety layer: ✓ Red flag detection (chest pain, suicidal ideation, acute trauma) ✓ 911 escalation — not "call your doctor," not "see a practitioner," actual 911 ✓ Ephemeral conversations — no PHI stored on server, ever Helpful without being medical. Recommends without diagnosing. Escalates when it should.
The Karma Economy
Double-entry ledger for practitioner karma points. Vendor offers, referral bonuses, community currency that incentivizes collaboration over competition.
“I had HealFest in five weeks and needed a full marketplace — practitioner directory, AI wellness guide, karma economy, payments, the works. Travis and The Collective shipped the entire platform before the event. Then they migrated seven years of blog content from WordPress without losing a post. I went from ‘I don’t know if this is even possible’ to a production product with real practitioners and real traffic.”— Reena Jadhav · Founder, HealCircle
What AI made possible
Without AI, HealCircle is a directory with a search bar. With it, consumers get a guided conversation that ends with matched practitioners, an action plan, and — if symptoms are serious — direct 911 escalation. The AI transforms a marketplace into a care navigator.
Ignite Leader
An AI-native leadership simulation platform where an adaptive coaching agent runs learners through 8 modules of real-world scenarios — not courseware, not video lectures. Decisions have consequences, the agent remembers everything, and every session produces tangible deliverables.
The model
50/50 equity partnership — co-building, not billing. Full shared access to codebase, governance, and decision history. Both parties succeed when the product succeeds.
What we built
- 8-module curriculum engine — structured AI coaching conversations, not videos
- Claude-powered coaching agent with cross-module memory and adaptive responses
- Leadership Identity Map built through guided self-assessment
- AI Fluency Score — benchmarked assessment with scored reports
- Capstone: AI Investment Pitch coached by the agent
- Dual-mode: "Teach Me to Fish" (deep) vs "Give Me the Fish" (accelerated)
- Per-module downloadable deliverables
- Client dashboard — cohort management and progress tracking
Stack
What a session looks like
Module 3 — AI Inventory & Reframe Agent: "You mentioned your team uses Slack, Google Docs, and a project tracker. Let's map what you're already doing with AI — even if you don't call it that." Learner gives input... Agent: "Three of those five tools have AI features you're not using. Before I show you which ones — what's one task this week that felt repetitive?" Learner responds... Agent: [Builds personalized AI tool map] [References Module 1 leadership style] [Produces downloadable deliverable] Memory carries forward: Module 4 references this tool map Module 6 capstone builds on it Agent never asks the same question twice
The 8-module curriculum
Mod 1 Leadership Identity Foundation Mod 2 AI Readiness Assessment Mod 3 AI Inventory & Reframe Mod 4 Custom AI Tool Building Mod 5 60-Day Strategic Focus Mod 6 Capstone: AI Investment Pitch Mod 7 30-Day Reality Check-in Mod 8 90-Day Outcomes & Alumni Network Each module: ✓ AI coaching agent conversation ✓ Downloadable deliverable produced ✓ Cross-module memory preserved ✓ Progress tracked per learner
Why this model exists
Hourly billing breaks down when the product is undefined. Equity partnership aligns incentives — same governance, shared as co-owners instead of delivered as reporting.
What AI makes possible
Without AI, this is choose-your-own-adventure with a fixed script. With AI, every module responds to the learner's actual decisions. The agent references what they said three modules ago and produces deliverables unique to their organization. Not studying leadership — practicing it, with a coach that remembers everything.