AI Systems I Have Designed and Built
FairInfer LLC is available for AI systems consulting, discovery, prototyping, integration, and implementation. The projects below show the systems engineering, product development, and AI integration work behind that practice.
ClairaVoice
Private, local-first AI voice companion research and development
Founder-led FairInfer LLC R&D
ClairaVoice is not currently being offered as a commercial product. I am seeking an incubator, fellowship, or similar program with salary support, mentorship, and structured development time to continue development, prepare for pilot testing, and explore a sustainable product path.
Older adults living alone face real risk from isolation and cognitive decline, but voice assistants built on cloud AI ask them to trade privacy for help, sending household conversations to a third party with no clear boundary on what happens to that data. ClairaVoice is an ongoing FairInfer LLC research and development project testing whether a voice companion can offer memory and practical assistance without that tradeoff. The current system is a working technical prototype focused on validating architecture, privacy, interaction design, reliability, and the path toward real-world pilot testing.
Its core AI stack runs locally instead of depending on cloud AI APIs. AI inference is separated onto NVIDIA Jetson hardware while voice services, persistent semantic memory, workflows, backup/recovery, and controlled network access operate across local Linux systems.
Technical scope
- Local LLM inference and NVIDIA Jetson edge AI
- Voice AI architecture, speech-to-text, and text-to-speech
- Persistent semantic memory and Rust workflow orchestration
- Linux and Docker deployment under hardware constraints
- Private-network design, backup and recovery
- Rollback and acceptance testing
ATHENA
Traceable AI-assisted software delivery
AI coding agents move fast and forget everything. Without a record connecting what a customer asked for to what got built, teams lose the ability to explain why a piece of code exists, and every new agent session starts from zero context. ATHENA is an agent-agnostic framework that fixes this by preserving the connection between customer requests, decisions, requirements, implementation, and shipped code. It creates a traceable path from customer request to design decision to functional requirement to implementation task to execution evidence to Git commit.
A later version added a deterministic context gateway so agents can retrieve task-relevant information without repeatedly loading the project's full historical record, cutting the token cost of staying oriented.
63%
lower total context consumption in the Claude Code test
13.5%
lower total context consumption in the Codex test
~94%
lower fixed session-opening context in those tests
Technical scope
AI-assisted engineering workflows, requirements traceability, agent context management, deterministic tooling around LLMs, software delivery controls, and measurement and evaluation.
View ATHENA on GitHub open_in_newCORTEGE
Multi-agent security system prototype
Scam and fraud attempts increasingly target specific households with specific knowledge of who lives there, and generic fraud alerts don't account for that context. CORTEGE is an AI security companion prototype built around household-specific agents that assess scam, fraud, and security signals against what's normal for that household. The architecture treats the language model as one component inside a larger software system, with a React frontend, Express backend, REST and WebSocket APIs, agent orchestration, persistent state, encrypted personal information, and audit trails, rather than allowing the model itself to become the application.
Technical scope
Multi-agent architecture, agent orchestration, REST and WebSocket APIs, AI application integration, PII encryption and redaction, SQLite persistence, tamper-evident audit trails, and privacy-aware system design.
View CORTEGE on GitHub open_in_newAI Interview Practice
Document-driven realtime voice application
Generic mock-interview questions don't prepare candidates for the interview they're actually going to have, and most practice tools give feedback well after the moment it would have been useful. AI Interview Practice closes both gaps: it generates personalized interview sessions from a user's resume and target job description, evaluates written and spoken responses in realtime, provides structured feedback, and preserves sessions so a candidate can track improvement over time.
Technical scope
Realtime voice AI, WebRTC integration, document ingestion, structured LLM output validation, AI evaluation workflows, persistent application state, FastAPI development, and user-facing AI product design.
View AI Interview Practice on GitHub open_in_newAte the Mic
Real-time AI public-speaking coach
Most people don't get worse at public speaking from lack of knowledge, they get worse from lack of reps with honest feedback in the moment. Ate the Mic gives them that: a coach that watches and listens while they speak and responds in real time, not after the fact. Four modes cover different needs, guided practice for building a talk, composure drills for staying steady under pressure, timed challenges for pacing, and lower-pressure sessions for beginners, with session history so progress is visible over time.
Technical scope
Built on the Gemini Live API for native audio and bidirectional WebSockets, with browser mic/camera streaming, a React/TypeScript frontend, Express backend on Google Cloud Run, Firestore for session data, managed authentication, and automated unit and end-to-end testing.
Have an AI system or workflow problem?
FairInfer LLC works with organizations that have a real operational problem and need someone who can move from discovery into a working system without requiring them to build an internal AI team.
Available for discovery, architecture, prototyping, integration, implementation, and embedded AI systems work. Current consulting rate: $125/hour.
Email me arrow_forward