01 · FEATURED
AI systems in production.
Customers described by industry.
Platform components I architected at MoolAI, agent systems I shipped for enterprise customers, and my graduate research.
Judge Council
LLM-as-a-judge evaluation that checks agents before release and on live traffic.
ContextForge
Context engineering that cut agent token spend by about 30%.
4 days → 30 min
A financial review cycle rebuilt as a supervisor agent delegating to specialist sub-agents.
5× faster to care
Clinic voicemail triage that sorts calls by urgency and routes emergent patients to a nurse first.
−50% triage latency
Multi-agent alert triage on an isolated on-prem network at 99.9% uptime, delivered with a team of three.
7 days → 1 day
Reconciliation across supply-chain systems that had never talked to each other.
The model holds no approval authority
A deal-desk agent where a policy engine decides discounts and escalations go to a person in chat.
Per-seat → usage-based
Chat agents over planning, ERP, and accounting data, so business users work where they already are.
12 models, one leaderboard
An experiment engine that benchmarks a customer's agent across models and ranks them by quality and cost.
FogBrain
Private LLM inference on a Raspberry Pi cluster: sub-150ms responses, zero cloud, 25% less power.
02 · ARCHIVE
Earlier work.
Where the foundations came from.
Machine learning, analytics, and data projects from grad school, where I built my foundations in modeling and evaluation.
