Proven experience with system refactors and schema migrations
Familiarity with PostgreSQL and AWS
Experience in building production systems for AI/ML workflows
Comfortable working across the entire tech stack
Curious and pragmatic with a strong understanding of complex data flows
Responsibilities
Enhance AI knowledge base and retrieval systems for accuracy and relevance
Rebuild data pipelines for timely, trackable information
Develop observability tools for monitoring system performance
Ensure compliance during data migrations and transfers
Create systems for seamless data integration with health partners
Benefits
Work alongside a team of ex-founders and top engineers with proven success
Opportunity to shape a fundamental healthcare solution
Direct impact on patient care and healthcare accessibility
Collaborate with leading clinicians and researchers in the field
Chance to redefine the understanding and use of healthcare data
Full Job Description
Member of Technical Staff @ Lotus AI
What this role is
You'll help build and operate the AI + data systems behind AI-driven primary care.
This is a generalist role. You may work across model training and fine-tuning, model tooling, data pipelines, retrieval/evals, and product workflows.
You'll be close to the core system and involved in product decisions from day 1.
You'll design and scale the data and retrieval systems that power Lotus's clinical AI, improving correctness, traceability, and explainability in how medical information is surfaced, validated, and applied in real-world care.
You'll help shape our real-time voice and video AI capabilities, building the foundation for intelligent, multimodal patient interactions.
What this role is not
Not a big-company role with tight scope and clear lanes.
Not a place with a formal hierarchy or long onboarding ramp.
Not a "ticket queue" job. Priorities will change week to week based on user needs, clinician feedback, safety issues, and what's breaking.
What you'll do
AI Agents and Product Intelligence
Build and iterate on AI agent workflows that handle multi-step clinical reasoning, tool use, and structured decision-making.
Design guardrails, fallback logic, and escalation paths to ensure safe autonomous behavior in patient-facing products.
Prototype and ship new AI-powered product features end-to-end, from model selection to UX integration.
AI Knowledge Base and Search Improvements
Improve knowledge bases so that citations resolve to original data and searches are fast, relevant, and prioritize tier-one medical information.
Continuously enhance retrieval accuracy and data lineage tracking.
AI Data Ingestion and Integrity
Rebuild data pipelines to eliminate stale data, support clinician and patient corrections, and ensure full traceability.
Design models that sync cleanly with health data partners and credentialing authorities.
Build and maintain data curation pipelines that produce high-quality training and evaluation datasets from clinical interactions.
Voice and Video AI
Build and optimize real-time voice pipelines for patient-facing interactions, including speech-to-text, natural language understanding, and text-to-speech.
Develop low-latency, streaming voice agents that can conduct clinical intake, triage, and follow-up conversations with empathy and medical accuracy.
Fine-tune voice and video models for medical terminology, diverse accents, and accessibility needs.
Design interruption handling, turn-taking logic, and conversational state management for natural, fluid voice experiences.
Observability and Analytics
Build monitoring and analytics for background jobs to monitor failure rates and identify partner vs. internal issues.
Streamline tracing, logging, and auditing to reduce redundancy while maintaining compliance-grade visibility.
Instrument model performance tracking in production - monitoring latency, token usage, output quality, and drift over time.
Model Training and Fine-Tuning
Fine-tune and adapt foundation models on clinical data to improve diagnostic accuracy, safety, and tone for patient-facing interactions.
Design and run training pipelines including data curation, annotation workflows, hyperparameter tuning, and model evaluation.
Develop and maintain evaluation frameworks (automated and human-in-the-loop) to measure model quality, safety, and regression across releases.
Experiment with prompt engineering, RLHF, distillation, and other techniques to optimize model behavior for healthcare-specific use cases.
What you bring
Strong programming skills, preferably Python
Experience with system refactors, schema migrations, and data infrastructure simplification
Familiarity with PostgreSQL (including JSONB and vector types) and AWS
Experience building production systems that power AI or ML workflows
Hands-on experience with LLM APIs, prompt engineering, and shipping AI-powered product features
Comfort working across the stack, from schema design to production debugging
Bonus points
Familiarity with training infrastructure and frameworks (PyTorch, Hugging Face, vLLM, Axolotl, or similar)
Experience with RLHF, DPO, or other alignment and preference-tuning techniques
Experience building or improving AI agent systems with tool use and multi-step reasoning
Experience building retrieval systems for LLMs (RAG pipelines, vector search, grounding)
Familiarity with FastAPI, SQLAlchemy, DuckDB, Temporal, ClickHouse, Valkey, or similar systems
Experience with real-time voice AI systems, speech models, computer vision, medical imaging, or multimodal models that combine text, audio, and visual inputs
Knowledge of logging/monitoring stacks (Sentry, Langfuse) and containerized deployments (Docker, ECS)
Experience simplifying multi-layered data systems where architectural issues cascade through storage, logging, and application layers
Strong intuition for designing systems that balance correctness, observability, and performance
What success will look like in the first 90 days
30 days
Shipping reliably, understands the core system, owns a small surface area
60 days
Owning a meaningful system and improves a key metric (quality, latency, clinician wait time, data reliability, etc.)
90 days
Independently driving a roadmap slice and raises the team's bar (agents/evals/monitoring)
