I ship AI systems that actually get used — specifying what to build, driving implementation with engineering, training the people, and measuring what sticks. I've done it in legal operations, clinical health tech, and for fun.
End-to-end implementations, AI deployments, adoption systems, and things I built just because I wanted to.
Built the West Coast CS function from scratch — hired a 7-person team, defined the org structure, and standardized implementation workflows with enterprise-grade SOPs. Scaled the portfolio from 29 → 65 facilities across 8 states in 12 months while cutting onboarding time in half and recovering lost device revenue.
Led a high-velocity enterprise rollout across 72 facilities in 3 states over 5 months — running 10–20 parallel implementations per month. Built the full infrastructure from scratch: SOPs, launch playbooks, and role-specific clinical training for physicians, nurses, and facility leadership. Conducted on-site workflow assessments to identify friction, then transitioned accounts to self-sufficient models through structured virtual follow-ups.
Turned high-engagement clinical users into a peer-driven growth channel — building a structured ambassador network with regional events, QAPI case study presentations, and a "Circadia Certification" program. Converted 15 of 26 pilot facilities to paid through ambassador-led referrals and internal advocacy across NorCal and SoCal.
Built a 0–100 weighted health score per facility using Mixpanel + Preset, tracking 7 feature depth metrics, DAU/WAU, session time, ticket volume, and NPS. Automated 6-tier alert logic to shift the CS team from reactive check-ins to proactive intervention — adopted as the single source of truth across the full portfolio.
Replaced a manual QA nurse escalation workflow with an LLM-powered in-app system that analyzes structured and unstructured patient data, generates escalation reports, scores deterioration risk, and surfaces clinical recommendations in real time — all in a HIPAA-compliant environment. Designed the clinical UI in collaboration with engineering, piloted with ambassador clinics, and managed the QA nurse transition to the automated system.
Replaced a manual intern-run QA reporting process with a live in-app dashboard — integrating EHR data, tracking intervention outcomes via GPT analysis, and calculating per-resident cost savings in real time. Gave facility executives and CS teams the data to drive renewal and expansion conversations across 85+ facilities.
Built and trained an automated friction detection system in Intercom that identified behavioral stress signals in real time and surfaced proactive, contextual help before clinical users disengaged. Shifted the support model from reactive ticket-handling to in-the-moment intervention — reducing volume while improving adoption and feature stickiness.
Co-developed a proof-of-concept AI agent with the CEO and engineering team to automate clinical intake at skilled nursing facilities — parsing 30–100+ page discharge PDFs via Amazon Textract, flagging contraindications, and generating CMS-aligned draft care plans. Eliminated 1–2+ hours of manual nurse review per patient in a HIPAA-compliant environment.
Replaced a 17-person manual document processing team with an OCR + LLM pipeline (Claude Opus) that splits large PDF productions, classifies documents by type and subtype, and extracts structured metadata for instant retrieval. Managed 2 data engineers through build, QA, and validation — mapping the full firm document taxonomy with litigation SMEs and building sampling thresholds to handle edge cases at scale.
Built a centralized AI drafting studio for trial prep — pairing a firm-approved template library with an LLM layer that generates case-specific content from indexed documents. Shifted attorneys from drafting to reviewing. Reached production quality with 3 attorney end-users in under 2 months.
Identified a manual research bottleneck in lemon law litigation and built a Claude-powered tool that automatically retrieves and organizes all relevant NHTSA filings, complaints, investigations, and recalls for each case — validated with SMEs and integrated into the firm's document management system.
Designed an AI decision-support agent that eliminated ~16 hrs/week of manual spreadsheet work — pulling structured case data from Palantir and generating LLM-scored recommendations across 23 data points to guide attorney settlement decisions. Built to augment judgment, not replace it, with transparent scoring logic attorneys could interrogate and override.
As caseload grew 23%, inbound client calls were overwhelming the team. Designed Salesforce SMS automation with staged workflows mapped to each litigation phase, enriched the client portal with educational content and checklists, and redesigned two core ops functions around AI — cutting call volume while the firm kept scaling.
I build AI agents in my personal time — not for work, just because I find the problems interesting. These are three that are live and running.
Built to solve a real problem — daily decision fatigue around food. I mapped the logic first (restaurant pool, recent meal history, macros, Uber Eats link), then used Claude as a thought partner and technical teacher to write the Python that powers it. I can't write code from scratch, but I can read it, debug it, and direct an AI to build it — and this is the proof.
Claude API · Python · Twilio · Make · JSON · Cron
Generic apps give you a notification — I wanted something that gives you context. Built a daily coaching agent that texts the morning workout, intensity zone, fueling plan, hydration targets, and training block context for a 16-week Ironman 70.3 plan. The agent reads the full plan from JSON, finds today's session, and generates the message — so it knows if it's a recovery week, not just what day it is.
Claude API · Python · Twilio · JSON · Cron
I sit at the intersection of AI strategy, customer success, and business operations. My background spans healthcare tech, legal AI, and enterprise SaaS — which means I know how to design and implement systems that real people actually adopt in high-stakes environments.
I've trained nurses, attorneys, facility CEOs, and engineers. I've built the playbooks, the dashboards, the training programs, and the prompts. I care about what happens after the tool gets deployed — and I build things on my own time to stay sharp.
UCLA Neuroscience, Summa Cum Laude. Ironman 70.3 in training. Based in Los Angeles.
What the people I've worked alongside have to say.
"I had the pleasure of working with Beylem and highly recommend her. She is reliable, proactive, and brings great attention to detail to everything she does. Beylem is a strong team player who consistently goes above and beyond. Any company would be lucky to have her."
"I reported to Beylem at Circadia Health on the Customer Success team. She is a warm, energetic, and thoughtful leader who brings real value to any team, regardless of the industry. She is so supportive while still delivering clear, constructive feedback. She's consistently proactive and creates an environment where you're pushed to improve while feeling backed. Beylem is also incredibly forward-thinking — always looking ahead, anticipating what's needed, and thinking about how to scale in a way that sets the team up for long-term success."