AI products.Built. Scaled.
I build production AI products and the infrastructure that makes them reliable — from multimodal apps and recommendation agents to evaluation gates for high-stakes LLM systems.
Outcomes, not adjectives.
AI systems with public proof.
Evaluation infrastructure, multimodal products, retrieval agents, and production systems with real users and adoption.
PrimitiveBench
Vendor-neutral evaluation infrastructure for selecting and validating the primitives behind production AI systems.
NutrioMN
A Mongolian-first nutrition app using a fine-tuned vision-language model to turn food photos into reliable meal logs.
VAULT Collection OS
A full-stack operating system for tracking, valuing, and understanding high-value physical collections.
CourseLynx
Course discovery agents, catalog ingestion, and Chrome extension infrastructure serving students across 16+ universities.
Curio AI
A LangGraph recommendation system combining live retrieval, structured outputs, and persistent learner progress.
NeuronBook
A grounded PDF learning system with Socratic questioning, concept memory, and a visual knowledge graph.
PM.ai
A multi-agent workflow that turns unstructured project briefs into validated tasks and database-backed execution state.
Google Ads Transparency Monitor
An Apify monitoring primitive that converts public advertising data into repeatable, automation-ready datasets.
I care about the version of AI that survives outside the demo — where real users, real money, and real failure modes are on the line.
As a technical founder I've shipped a Mongolian-first multimodal nutrition app to 2,500+ users, co-founded PrimitiveBench — a vendor-neutral evaluation layer for AI infrastructure now partnered with CalCompute — and built recommendation and retrieval systems serving students across 16+ universities.
The throughline is discipline around the model: structured outputs, golden datasets, agent traces, deterministic fallbacks, and CI eval gates. Flexible reasoning becomes a reliable product only when its output is measurable, debuggable, and safe downstream.
I move at founder speed but build with engineering restraint — because taste, distribution, and reliability are all part of systems design.
Technical depth, made legible.
Core patterns behind measurable, debuggable agentic systems.
Production AI products
From prototype to paying users, uptime, subscriptions, and growth.
Evaluation infrastructure
Golden datasets, agent traces, structured validators, and deployment gates.
Multimodal intelligence
Fine-tuned vision-language models with reliable structured outputs.
Agentic recommendations
Tool calling, retrieval, ranking, and live external data integrations.
Founder speed, engineering discipline.
Production ownership inside small teams and real constraints.
Founder & CTO
Shipped a Mongolian-first AI nutrition app, received an angel-funded grant, scaled to 2,500+ users, and led interns across product and growth.
Co-founder & Lead Engineer
Built an open-source, vendor-neutral evaluation layer and converted PrimitiveBench evaluation reports into $5K MRR from early customers.
Software Engineering Intern
Built real-time recommendation and retrieval systems serving students across 16+ universities.
Software Engineering Intern
Developed APIs, structured recommendation logic, and a LangGraph agent for dynamic content retrieval.
A practical technical range.
Tools for connecting model quality, system reliability, and product delivery.
AI / LLMOps
- Agent orchestration
- RAG evaluation
- Structured outputs
- VLMs
- Vector search
- Fine-tuning
Frontend
- React / Next.js
- TypeScript
- JavaScript
- TailwindCSS
- Flutter / Dart
Backend
- Python
- FastAPI
- Node.js
- REST APIs
- JSON Schema
Databases
- Postgres / SQL
- Supabase
- Firebase
- Redis
- Chroma
- Meilisearch
Infra / Tooling
- Docker
- GitHub Actions
- CI/CD
- OpenTelemetry
- Analytics
- A/B testing
Building something that needs to work outside the demo?
Open to AI systems, LLMOps, backend, and startup engineering opportunities. Email me directly, or verify the work through LinkedIn, GitHub, and PrimitiveBench.