Wilson Gichuhi, Systems Engineer
Systems engineer building reliable AI, integration, and infrastructure software for healthcare and commercial teams where decisions need context, traceability, and human accountability.
An AI-native platform that transforms fragmented, unstructured healthcare documents (images, referrals, scanned PDFs) into a grounded knowledge graph using vector embeddings and relationship-native reasoning. Automates provider-payer loops with low-latency voice agents and human-in-the-loop validation, building institutional memory that reduces referral delays and claim denials.
A 24/7 AI agent that joins video calls to deliver multilingual, interactive product demos. Drops into any SaaS product to navigate it via VLM-based agentic automation, qualifying leads and booking follow-ups directly in your CRM and calendar. Built on LiveKit, Deepgram, and Cartesia (<700ms perceived latency).
An event-sourced control plane for AI behavior. It versions context, decisions, and human corrections; turns approved learning into portable commits; and causally replays historical decisions across interchangeable runtimes.
Sept 2026 - Present
Co-founding CareWeave, a provider-side revenue integrity system for insured care. Building the workflow and reasoning infrastructure that helps hospital teams resolve financing exceptions across HMIS platforms, payer workflows, documents, and human follow-up. Designing event-sourced operational workflows, deterministic rule evaluation, context graphs, and governed human-in-the-loop AI for high-trust healthcare operations.
2026
Building Promco, the commercial decision layer for beverage distributors. It captures operational judgment between conversation and commitment, turning WhatsApp orders into commercially reviewable decisions before they reach ERP or downstream systems. Designing reliable AI systems, durable workflows, and decision infrastructure for human-in-the-loop commercial operations.
5 months
Contributed task authorship to OpenThoughts, Bespoke Labs' open reasoning dataset initiative, producing curated training data for small reasoning models in collaboration with Stanford, Berkeley, and UT Austin. Designed and authored Terminal Bench style agentic coding tasks in the Harbor format, covering concurrency bugs, semaphore leaks, and write-ordering failures across real-world TypeScript and healthcare system scenarios. Owned the full pipeline from environment setup through oracle verification. Applied Bespoke-MiniCheck for claim-level factual grounding in evaluation pipelines, using NLI-based verification alongside LLM-as-judge methods to improve signal reliability in automated test suites.
2025
Built evaluation environments for frontier models across infrastructure, security, and healthcare workflows.
2025
Worked on a multi-tenant FMCG operations backend covering stock intelligence, payments, dispatch, and promo tracking.
2024 - 2025
Architected a secure telemedicine platform across triage, pharmacy, and lab integration using WebRTC and local payment gateways.
2024
Built a CRM voice-logging assistant and rebuilt a legacy PHP marketplace into an offline-first Next.js application.