LOCAL_INFERENCE / AGENT_SYSTEMS / FAILURE_RECOVERY
AI/ML engineer building local-first systems, practical automation, and tools around real problems
currently building → Hermes Agent + custom tool harness
Important
Open to AI/ML internships around local inference, agent reliability, industrial ML, and embedded / edge systems.
|
PUBLISHED PATENTS 03 |
EDGE INFERENCE <100 ms |
HACKATHON Winner |
|
LOCAL INFERENCE / PRODUCT SYSTEMS Local-first Windows dictation with packaged speech-model workflows, global hotkeys, a click-through live overlay, searchable history, automatic insertion, CPU/CUDA fallback, and hardened WebSocket session transitions.
Why it matters: the interesting work is not merely transcription — it is getting inference, desktop runtime, session state, model delivery, and release packaging to behave as one product. |
INDUSTRIAL ML / STREAMING Kafka, Spark, Cassandra, PyTorch, and graph analytics across 5,000 simulated factory machines, combining streaming anomaly detection with cascade-failure analysis and constrained mitigation.
Why it matters: model output is connected to a live data path and downstream decisions rather than treated as an isolated notebook prediction. Repository private · available on request |
|
UNSUPERVISED ML / RESEARCH HDBSCAN proposes operating regimes on NASA CMAPSS and CWRU while domain constraints reject mechanically or thermodynamically implausible clusters.
Why it matters: discovery is not accepted simply because a clustering metric likes it — the result also has to make physical sense. |
EMBEDDED INFERENCE / RASPBERRY PI Medication verification using quantized MobileNetV2 on Raspberry Pi 4. Low-confidence predictions route to a human fallback instead of pretending certainty.
Patent: |
Most of my projects start with friction I run into myself.
If something is repetitive, awkward, slow, or keeps getting in the way, my first instinct is usually to ask whether I can automate it, simplify it, or build a small tool around it. Sometimes that becomes a larger project; sometimes the useful result is just one focused fix.
The same applies to problems I see other people repeatedly dealing with. I like understanding the workflow first, then building around the actual pain point instead of starting with a technology and looking for somewhere to use it.
I also spend time digging through open-source projects — using them, reading how they are put together, borrowing good ideas, and figuring out what I would change for my own workflow.
Problem first. Tool second. Keep what actually makes the workflow better.
| Track | Systems |
|---|---|
| AI / Product | Eve · ZRead · Stella |
| Agents / Systems | Cognitive Load Scheduler · Aether Dashboard |
| Edge / IoT | Vision Air Sim · Air Safety Assistant · Smart Energy |
| Vision / Security | Face Privacy Filter · Sobel CUDA · QR Security · SecureTorrent |
AI is as much an interest for me as it is an engineering tool. I follow new model releases, read model cards and benchmarks, test them against real tasks, and work out where they actually fit into my workflows.
I am especially interested in agentic systems, coding agents, tool use, orchestration, inference, and the rapidly changing model landscape. When something new ships, I want to understand what changed, what it is genuinely good at, and whether it changes the way I build.
New models are interesting. New capabilities that change a workflow are much more interesting.
|
Experience Brandworks Technologies · AI/ML Intern |
Hackathons Hackathon Winner · IEEE CS 2025 |
|
Patents
Secure sensing · embedded medication verification · cognitive-load-aware distributed scheduling. |
Credentials IBM Agentic AI · NVIDIA DLI Deep Learning · Oracle OCI AI Foundations. |
notice the friction · build the fix · keep what works