I build AI systems that know when to answer, abstain, escalate, retrieve, or recover.
| Signal | Evidence |
|---|---|
| Original AI thinking | Metacognition benchmarks, bounded inference, abstain/escalate behavior, hidden-state reasoning. |
| Applied ML range | RAG, OCR, point-cloud registration, anomaly detection, reinforcement learning, document AI. |
| Product ability | React, TypeScript, Node, Express, FastAPI, MongoDB, dashboards, APIs, deployment surfaces. |
| Communication | Every public project now has a description, README context, topics, and a clearer role in the portfolio. |
| Clickability | 15 repositories now have live/demo/homepage surfaces so reviewers can inspect before cloning. |
| Project | What it proves | Code | Demo |
|---|---|---|---|
| Holup Benchmark Lab | Evaluation design for commit vs abstain vs escalate behavior in LLMs. | Repo | Open |
| DEIC Executive Inference | Hidden-state belief tracking, source reliability, query budgeting, and recovery. | Repo | Open |
| Kathakaar | Cultural AI with grounding, narrative generation, search, maps, and voice direction. | Repo | Open |
| Cons.trukt | Construction document intelligence, risk analysis, physics logic, and audit trail thinking. | Repo | Open |
| Healing Stones | AI-assisted 3D artifact reconstruction using point clouds and geometric alignment. | Repo | Open |
| Sonish v2 | Full-stack MERN commerce app with frontend, backend, security baseline, and deployment path. | Repo | Open |
| Area | Projects |
|---|---|
| Metacognition and reasoning | Holup, DEIC, SAAMC |
| Applied AI products | Kathakaar, Cons.trukt, SketchAI, AppleO |
| ML and data systems | Water.DRIE, Net.AI, threat_dat, clearWattson |
| Product engineering | sonish-v2, Sid-s_terminal, accelerated_framework |
Python PyTorch FastAPI Jupyter React TypeScript JavaScript Node.js Express MongoDB OCR RAG Computer Vision Reinforcement Learning Cybersecurity
| Track | Next visible upgrade |
|---|---|
| AI evaluation | Public leaderboards, richer model comparison dashboards, downloadable benchmark artifacts. |
| AI demos | Hugging Face Spaces / Streamlit / Gradio interfaces for notebook-heavy ML projects. |
| Full-stack systems | Hosted frontend + backend demos with seeded data and demo credentials. |
| Portfolio polish | Screenshots, architecture diagrams, result cards, and tighter README storytelling. |
AI/ML internships, research engineering, applied LLM evaluation, agent systems, public-interest AI, and product engineering roles where prototypes need to become real tools.