AINPM.in / Work
Portfolio · Updated 16 Nov 2026

Products I've built and shipped

Five open-source products by Ian Pinto: an AI assistant for SEBI regulation, a zero-knowledge password manager, a passkey cryptography library with a research preprint, a local-AI study app for macOS, and fine-tuned code search. Each card shows dated numbers. Open a case study to see the trust design and trade-offs.

5open-source products
1DOI-registered preprint
29skill-to-project connections
49public evidence links
01 · Products

What does each product prove?

Bars show measured rates; orange marks a known weak spot. Numbers are as of the dates on each case study.

Why does every product publish its numbers?

Because claims about AI are cheap and verified results are rare. Each product here ships with a dated evaluation, so a hiring manager or client can check the work instead of trusting a badge.

02 · Flagship results

How good is the flagship product?

SEBI Circular RAG has to clear six quality floors before any change ships. The tick on each bar is the floor; the bar is the measured result.

Learn more in the SEBI RAG case study

Quality on 260 adjudicated questionsFull evaluation 15 Aug 2026 · gate floors derived 2 Sep 2026
  • Abstention accuracy0.981
  • Recall@100.943
  • Context recall0.916
  • Citation recall0.881
  • nDCG@100.697
  • Citation precision0.194

Tick = release floor. Every metric clears its floor; citation precision is the known weak spot. Source: Evaluation log ↗

03 · Skills

Which skills does the portfolio prove?

Every line is a public evidence link from the skills matrix. Select a skill to see its proof.

Which projects prove which skills?29 connections · hover or focus a skill or project to trace it
Map of eight skills on the left connected to the projects, writing and credentials that provide evidence for them, on the right. The same information is listed below the chart.AI evaluation designRetrieval (RAG) product designTrust and compliance by design (India)Privacy and security architectureOn-device and local AIAI product strategy and MVP scopingTechnical program deliveryCommunicating AI quality honestlySEBI Circular RAG5 skillsTrustVault2 skillswebauthn-prf-zktv2 skillsLecture Lens3 skillsCodeSearchNet4 skillsOpen-source contributions4 skillsArticles & guides5 skillsCredentials4 skills
Show as a list
  • SEBI Circular RAG: AI evaluation design, Retrieval (RAG) product design, Trust and compliance by design (India), AI product strategy and MVP scoping, Communicating AI quality honestly
  • TrustVault: Privacy and security architecture, AI product strategy and MVP scoping
  • webauthn-prf-zktv: Privacy and security architecture, AI product strategy and MVP scoping
  • Lecture Lens: AI evaluation design, On-device and local AI, AI product strategy and MVP scoping
  • CodeSearchNet: AI evaluation design, Retrieval (RAG) product design, AI product strategy and MVP scoping, Communicating AI quality honestly
  • Open-source contributions: AI evaluation design, Privacy and security architecture, Technical program delivery, Communicating AI quality honestly
  • Articles & guides: AI evaluation design, Trust and compliance by design (India), On-device and local AI, AI product strategy and MVP scoping, Communicating AI quality honestly
  • Credentials: AI evaluation design, Retrieval (RAG) product design, AI product strategy and MVP scoping, Technical program delivery
Product artifacts

Where is the product work behind these projects?

Problem briefs, evaluation specs, metrics trees, decision logs and launch notes, each linked to the project's own documents.

Browse the product artifacts · See the dated timeline

04 · Open source

What have I contributed to other open-source projects?

Merged pull requests to projects I don't own, and small packages I publish and maintain. Only merged pull requests are listed. Download counts are small and stated as they are; checked 5 Nov 2026.

Merged pull requestProjectMergedWhat changed
Fix #563: prevent rationale-node leakage and preserve calls direction ↗Graphify-Labs/graphify2 May 2026Fixed two bugs that inflated node importance and flipped call directions in a code knowledge-graph tool used with AI coding assistants. Phantom edges in the repro fell from 7 to 0, with 5 new regression tests (441 to 446 passing).
Add inline variant to the Filters component ↗o2sdev/openselfservice10 Dec 2025Added an inline layout option to a customer-portal UI component, backward compatible with the drawer layout and documented with Storybook stories.
PackageRegistryFirst publishedWhat it doesRecent downloads
express-swagger-auto ↗npm22 Nov 2025Generates OpenAPI 3.x specs and Swagger UI from Express APIs via decorators, JSDoc and runtime capture. v0.3.2.45 npm downloads, 5 Oct to 3 Nov 2026
chroncraft ↗npm27 Nov 2025Lightweight, tree-shakeable date and time library with time-zone handling. v0.3.0.20 npm downloads, 5 Oct to 3 Nov 2026
express-middleware-chain ↗npm7 Dec 2025Fluent, typed middleware composition for Express with validation, rate limiting and caching. v0.1.0.16 npm downloads, 5 Oct to 3 Nov 2026
password-suite ↗npm27 Nov 2025Cryptographically secure password and passphrase generation with strength analysis, on the Web Crypto API. v2.0.1.14 npm downloads, 5 Oct to 3 Nov 2026
express-type-safe-routes ↗npm7 Dec 2025Type-safe Express routes with Zod inference for params, query, body and responses. v0.1.0.8 npm downloads, 5 Oct to 3 Nov 2026
express-env-validator ↗npm8 Dec 2025Fail-fast, typed environment configuration checks for Express apps. v0.1.0.7 npm downloads, 5 Oct to 3 Nov 2026
pandera-unified-validator ↗PyPI28 Nov 2025Unifies Pydantic record validation with Pandera DataFrame checks across pandas and Polars. v0.1.0.11 PyPI downloads in the month to 5 Nov 2026

All 5 case studies above and every package are on GitHub ↗ and Hugging Face ↗. A machine-readable version is at /portfolio.json.

05 · Experiments

What am I experimenting with?

Early public experiments, listed with their real status. None of them claims results it has not published.

ExperimentStartedLast activityWhat it exploresStatus
MLX AutoResearch agent ↗4 Aug 20265 Aug 2026A local agent (Qwen on MLX) edits one training script, trains for 5 minutes on an Apple silicon Mac, and keeps the change only if validation loss improves.Harness and loop spec published; no results yet
SEBI regulatory knowledge-base agent ↗27 Nov 202528 Nov 2025An agentic RAG toolkit over SEBI regulations in LangChain.js and LangGraph; an early step before SEBI Circular RAG.Prototype, superseded
UPI offline payment PWA (NFC) ↗15 Nov 202528 Nov 2025Design for offline UPI payments by NFC tap with QR fallback, enforcing UPI Lite limits (Rs 1,000 per payment, Rs 5,000 balance).Scaffolding only; payment flows not built
GenAI Tasker Lite (Android) ↗13 Sept 202531 May 2026An offline-first Android 14 task manager with on-device semantic search, plus a phased roadmap to real embeddings.Educational scaffold
06 · Common threads

What do these projects have in common?

Evidence before claims

Every project ships with tests or evaluations, and publishes its weak numbers alongside the strong ones.

Privacy by architecture

Keys stay on the device, models run locally, and storage alone is never enough to read user data.

Decisions written down

Rejected ideas, threat models and architecture decision records are part of the product, not an afterthought.