My Betaworks Application: Building AI That Reduces Work for Doctors
One-liner
FindUrMeds is our first wedge into building software that automates tedious tasks for doctors. To solve medicine shortages — 300 medicines in shortage according to FDA — we cold-call dozens of pharmacies per patient until we find their medicine, and then share with patient and provider. We are next focusing on prior auth.
Describe your product
There are two parts to our initial wedge: a GoodRx-style user experience with both provider and patient portals to input medicine and patient information on, and the AI agent on the backend that makes the calls.
Doctors have money, little time, and are sick of tedious processes that make it harder for them to serve their patients. AI is a natural fit. One of my pilot providers told me his patients have been “raving about being able to find medication” — that’s the signal we’re solving a real problem.
Using these initial provider partnerships as buy-in, I’m expanding to prior authorization — a much larger market. Prior auth requires sending calls, faxes, and emails that are often ignored by insurance companies, just to get a patient’s medication covered.
The legacy player is CoverMyMeds. Doctors hate it. One psychiatrist told me: “I am not prescribing probably three quarters of medications because I know I don’t have time to do prior auth.”
CoverMyMeds won’t fix this — they’re a submission layer, not an intelligence layer. They don’t remember (and they easily could if they wanted) what medications a patient already tried, don’t reuse context across requests, and can’t be delegated to because questions come dynamically. They’re built for compliance, not for doctors. We’re building the AI layer that actually reduces the work.
Describe how your agent system works
We run three agent systems:
Lead Gen: Scrapes conference directories and practice websites. Uses LLM to piece fragmented contact info together — a name here, an email there, a phone number somewhere else. Qualifies and feeds into outreach.
Email Response: Classifies incoming replies (various states of interested, question, unsubscribe) and generates contextual follow-ups with Zoom links, custom documents, etc. autonomously. 5% meeting booking rate, with minimal human intervention.
Phone Agent (core product): For a given zip code, our agent creates a list of pharmacies likely to have the medication, calls them concurrently, navigates the “press 1 for pharmacy” menus, waits through hold, and then talks to staff — verifying details and checking if the medicine is in stock. 20-30 calls per search.
I run the ElevenLabs power user WhatsApp group and wrote a detailed breakdown of what it takes to build production phone agents: https://substack.com/home/post/p-183751053
Do you have any interesting strategies for distribution or growth?
Yes. I built an end-to-end automated pipeline: AI scraper extracts non-public psychiatrist contacts from fragmented medical sources → personalized cold email system → LLM agent classifies responses and generates contextual follow-ups. Currently at 5% meeting booking rate, targeting all US psychiatrists before expanding to pediatrics.
Cold email has been around forever, but sophisticated lead-gen systems paired with reply agents have not. This lets me compete against larger teams as a solo founder.
What’s your technical stack?
Voice AI: ElevenLabs for conversation + Twilio for telephony. ElevenLabs doesn’t work out of the box for real-world phone calls — I built significant custom infrastructure on top.
Non-exhaustive list of infra I’ve built (more detail in my Substack):
Chain-specific prompts and DTMF handling (CVS, Walgreens, Duane Reade all behave differently)
Custom analytics pipeline
Interceptor LLMs that catch poor responses before they’re spoken and orchestrate with the main LLM
Batch orchestration — 20-30 concurrent calls with isolated state per call
Natural language variation — so the agent doesn’t repeat itself and get flagged as a bot
Lead-gen and email systems: Custom Selenium scrapers + APIs for lead enrichment (SERP API) + LLMs that filter and classify leads. The email responder uses multiple separate LLMs as individual decision makers — one for classifying emails, others for navigating Google Calendar, Google Drive, Zoom APIs, etc.
This isn’t available in any lead-generation software I can find. Existing tools are too siloed and lack context awareness. For instance, no doctor wants to communicate using Calendly — it feels impersonal and adds friction. My agents recognize natural language like “can you talk in an hour” or “at 2PM,” convert it into an actual time (accounting for the doctor’s time zone), and reply quickly in their time zone. No annoying scheduling links.
Philosophy: I chain multiple Claude calls rather than relying on one prompt to do everything. More reliable, easier to debug.
AI-assisted coding: Cursor + Claude.
Who are your competitors and what makes you different?
MedFinder (medfinder.com) — direct competitor for medication shortages, similar approach: AI phone agents calling pharmacies. They’ve done ~$700k in revenue. But MedFinder has less contextual UX — they don’t help providers configure prescriptions or white-label the experience. We also ship features they don’t: medicine pick-up reminders and price-shopping to find the cheapest option. We’re focused on building provider relationships as a wedge into larger healthcare workflows — more profitable, and more impact on doctors’ lives.
CoverMyMeds (covermymeds.com) — the legacy player in prior authorization, and by far the dominant product on the market. There’s almost no competition, so little pressure to innovate. The UX is awful. Prior auth is essentially: figure out what documentation the insurance company wants (email? fax? phone?), assemble it from a maze of medical records, and submit — often repeatedly until they say yes.
Opportunity for disruption: CoverMyMeds helps (somewhat) with submission by letting you send an email. That’s about it. No gen AI drafts or templates. No EHR integration. It doesn’t remember what you’ve already sent or recommend what to do next. It should say: “You’ve sent 2 emails detailing the patient’s history. Based on our data, we recommend you next send an appeal to X — here’s a draft.” It does none of that. The through-line: LLMs + data will make the next generation of software far more contextual. Both of these products — CoverMyMeds especially — lack this entirely. That’s the gap.
What is another startup or founder that you admire? Tell us why.
Josh Miller and The Browser Company (Arc), pre-pivot.
Arc is design-centric and built intuitive, context-driven software with a reliable, relaxed feeling. Josh talks about building products where you can’t always articulate why it’s better, but you feel it. That resonates — if I’m going to win doctors, they need to feel comfortable relying on me and trust that I understand their POV.
Smart, DIY type distribution. They built in public, created genuine community, and let the product speak instead of spending on ads. Product as the marketing.
David vs. Goliath. They went up against Chrome with 65%+ market share and won users through craft. That’s similar to my position against CoverMyMeds — entrenched, dominant, zero pressure to innovate. You beat them by being meaningfully better on the things that matter.
Origin story & bios
I started FindUrMeds after getting fed up spending hours each month calling dozens of pharmacies to find my ADHD medication — and realizing there were thousands of other patients with the same problem. When I saw phone agents were starting to work, I realized that with the right custom infrastructure I could make them reliable enough to actually solve this.
Bio: I am a former software engineer at Google, PM at Instacart, early employee at Oscar Health. Solo founder based in Brooklyn.
Why are you the team to solve this problem?
As I tell doctors, I’m a patient first, business owner second.
I’m a solo founder with the full stack: a patient who knows the problem firsthand, an engineer who can build (Google), a PM who can design and prioritize (Instacart), and an Oscar Health alum with access to their founder community for mentorship. I’m deep in voice AI — I run the ElevenLabs power user WhatsApp group.
Three values guide how I build:
1. Design matters — where it counts. It’s easier to build software than ever, so the products that win are the ones that nail the details that actually matter. I’m not polishing every pixel — but when getting something wrong costs a patient hours of wasted calls (like selecting the wrong dosage), I invest the effort. I needed a medication database that didn’t exist, so I scraped GoodRx’s entire catalog myself. Not the most MVP-friendly choice, but as a patient I know how frustrating it is to accidentally select the wrong, similarly spelled medication. I built it right so users wouldn’t have to deal with that, and doctors appreciated it.
2. Scrappy beats polished. I got my first doctor customers at a medical conference — I paid for a guest pass (a booth was thousands of dollars), found an empty spot, and set up my own booth. After 2.5 days, the organizer politely asked me to stop — he said I was only the second person in 19 years to do this.
3. Product without distribution is a hobby. Solve the problem, then do whatever it (legally) takes to own the channel — even if it’s uncomfortable.
Try FindUrMeds: findurmeds.com
See a provider portal where doctors can white label their search for medicine: findurmeds.com/provider/malkin
Technical deep-dive on phone agents: read my Substack