Below is some of what I've been working on.
Distill An AI thought partner that turns raw notes, voice memos, documents, and web links into living, self-organizing wikis. Multi-agent synthesis with a tool-calling loop, an autonomous "dream cycle", and an LLM-as-judge eval harness to measure output quality. Vercel AI SDK · OpenRouter · Deepgram · LLM-as-judge evals · Next.js · TS → Live demo |
lead-nurture-pipeline AI SMS nurture system shipped for two detailing brands: 1,000+ messages a week integrated with their existing CRM, and a self-built eval loop that lifted weekly response rates from ~8% to a 29.5% peak. n8n · LLM · self-built eval loop · CRM integration → Case study |
blue-thumb-dashboard Stream-health dashboard on Google Cloud with an automatic daily sync and a Vertex AI assistant grounded in domain-specif context, visualizing 20+ years of Oklahoma water-quality data. GCP · Vertex AI · Dash · Python · ArcGIS → Live dashboard · Code |
kapwa-help Offline-first PWA for disaster-relief operations in the Philippines Offline-first PWA (Workbox · IndexedDB) · React · TS · Supabase → Live demo · Code |
specbook-extractor Turns messy Division 08 spec PDFs into clean, structured hardware data. A deterministic pipeline feeds an LLM extraction stage with forced structured output, audited at 97.1% field accuracy. Claude SDK · pdfplumber · OCR · Python · FastAPI → Live demo · Code |
realtime-interview-practice Voice-based AI interviewer: real-time spoken conversation over the OpenAI Realtime API, then an LLM grades the transcript against a rubric with structured output for scored, actionable feedback. OpenAI Realtime API · WebRTC · Next.js · TS → Code |
