Skip to content
View jwalin-shah's full-sized avatar
10 followers · 2 following

Achievements

Achievement: Pull Sharkx4Achievement: Pair Extraordinairex3Achievement: QuickdrawAchievement: YOLO

Achievements

Achievement: Pull Sharkx4Achievement: Pair Extraordinairex3Achievement: QuickdrawAchievement: YOLO

Block or report jwalin-shah

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
jwalin-shah/README.md
Jwalin Shah — Reliable agent systems, measured end to end.

I build systems that make probabilistic agents inspectable and dependable: isolated execution, deterministic verification, grounded retrieval, evaluation infrastructure, and fully local inference.

Portfolio · Resume · LinkedIn · Email

Systems and evidence

1. OpenHuman — Active · upstream
Personal memory and ingest in the Rust core: hybrid FTS5 + sqlite-vec retrieval, staged iMessage/Gmail/Calendar/Contacts ingest, contact identity via CNContactStore, and production hardening around citations, RPC, privacy/security, and chat reliability. [69 merged upstream PRs]

2. voice-engine-swift — Active · local product
Fully local dictation (Moonshine + CoreML). Bench: 26 ASR configs · 560 clips · moonshine-tiny 0.2177 WER / 0.05 RTF vs canary-qwen-2.5B 0.2162 / 0.73 — same accuracy class, 14× lower RTF. Tests: 62/62 v0.1 acceptance · 348/349 full regression (one CoreML OS regression documented).

3. officeqa-arena — Completed case study
Sentient Cohort 0 grounded QA: 184.5/246 (75%) at $1.71 · ~4,400 evaluations. Evidence selection caused 48% of failures; correctly grounded Python had 0% arithmetic errors. research.pdf.

4. Bridge — Private architecture · sanitized overview + runnable lease slice →
Spawn → verify → deliver for coding agents: deny-default sandboxing, leased worktrees, fresh-checkout verification. Public repo ships one inspectable assurance slice (TLA+ + Go property test + CI); full catalog stays private.

Capability strip: Rust+Go production core, Swift+CoreML local ASR, TLA+/Z3/Lean formal assurance, robotics ops across 5 platforms.

Background
OpenHuman (tinyhumansai) Core contributor — personal memory, ingest, hybrid retrieval
Break the Web / LiveLM AI Systems Engineer — grounded retrieval, MCP, ChatGPT path, eval harnesses
Skild AI Data Operations Lead — 5 robotic platforms · eval/ops systems · Series C demos

Contact

Open to agent reliability, evaluation infrastructure, and local-first systems roles.
✉️ jwalinshah13@gmail.com · 🌐 portfolio · 💼 linkedin

Pinned Loading

  1. tinyhumansai/openhuman tinyhumansai/openhuman Public

    Your Personal AI super intelligence. A brain that builds a local-first memory of your life, a fantastic orchestrator of agent fleets and workflows, and a deep researcher.

    Rust 36.1k 3.6k

  2. tensor-logic tensor-logic Public

    Working through Pedro Domingos' tensor logic paper — runnable demos walking from one einsum to continual learning.

    Go 1 1

  3. officeqa-arena officeqa-arena Public

    Competition entry: end-to-end OfficeQA pipeline for Sentient Arena — retrieval, ledger extraction, and LLM reasoning over 10k+ financial documents

    Python 3 1

  4. inbox inbox Public

    Unified inbox TUI: iMessage, Gmail, Calendar, Drive, Notes, Reminders, GitHub — local FastAPI server + Textual TUI

    Python

  5. data-connect-framework data-connect-framework Public

    Privacy-first personal data engine — schema-driven ingestion, canonical entity modeling, and local-first data pipelines

    Python