
Gaurav Narasimhan: Open Source AI Projects
An engineer’s portfolio of AI agents, document tools and research code
Updated November 16, 2026

Key facts
- What it is
- An engineer’s portfolio of AI agents, document tools and research code
- Pricing
- Free
- Category
- Tool & Resource Directories
- Website
- gaurav8936.github.io
What is this website for, in plain English?
Gaurav Narasimhan’s website is an engineer’s portfolio and a starting point for exploring his open source AI projects. Think of it as a map of software, research and current product work: each project explains a technical problem and points to its code or dedicated website. Developers can use it to find a component to inspect, a research implementation to study or an agent workflow to try.
The homepage itself is a public reading and discovery page. Running an agent, checking a document or training a model happens in the relevant linked project, with that project’s own setup requirements.
Coordinate coding agents with One Roof and roofd
One Roof brings sessions of Claude Code, Codex, Gemini CLI, Kimi and Antigravity into one macOS application. Its Inbox collects questions that need your decision; session history helps you catch up, and loops can repeat work against a check while leaving results for human approval. This is useful when planning, implementation and review span several agents.
roofd exposes the underlying session-management work. Its persistent terminal sessions are a separate engineering topic worth exploring if you are building agent infrastructure. One Roof’s desktop app requires Apple silicon, macOS 14 or later, git and an installed, signed-in agent CLI.
Check extracted facts and document changes
Three projects address different stages of document reliability:
- citation-verifier checks an extracted value against the document’s own text. Finding a genuine quotation and proving that it supports the reported number are separate checks.
- change-matcher aligns clauses across document versions and helps distinguish substantive changes from rewording, using retrieval, a cross-encoder and value-and-duty guards.
- docket-diff compares sections, recognizes renumbering and classifies regulatory changes with rule-based frames. Moving a section need not appear as an unrelated deletion and insertion.
These are components to inspect and integrate, rather than an upload-and-analyze service on the portfolio homepage.
Study agent memory and reliable model calls
llm-recall-probes tests whether a long conversation yields the right remembered value, an appropriate refusal or a misleading answer. It distinguishes fabrication from borrowing a real value from the wrong source and from accepting an incorrect number supplied by a user.
fail-open-scorer tackles service reliability in TypeScript: timeouts, response validation and a circuit breaker allow the caller to continue without model scores when scoring fails. It publishes the client, not the entire recommendation system or its models.
Explore code generation and clinical NLP research
FLeX investigates cross-language code generation with Code Llama-7B, LoRA and frequency-domain regularization. The MediQA 2023 repository contains work on classifying and summarizing clinical conversations with RoBERTa and BART. Read these as research implementations with particular datasets and evaluation conditions; they are not general accuracy guarantees or a clinical service.
From project card to a working local example
For a concrete first experiment, open citation-verifier from the portfolio, obtain its repository and read the README. With Python 3.11 or later, run these documented commands from the repository root:
python3 -m venv .venv
.venv/bin/pip install -e '.[test]'
.venv/bin/pytest
Then inspect the renewal-duration example. A document stating twelve months supports P1Y; an extraction of 24 months fails the value-support check even when its quotation exists. Compare those outcomes before adapting the library to your own fields. The optional PDF extra is needed for the demonstrated PDF bounding-box integration. Passing tests confirms your local setup, not universal correctness on new documents.
Access, licenses and practical limits
The portfolio was publicly readable when checked on November 5, 2026. One Roof is described as free and MIT open source; coding-agent accounts and optional hosted models remain separate services with their own charges. There is no shared model quota supplied by this website.
Several public repositories are selected copies of larger private systems. Prompts, datasets, routing or application pipelines may be excluded. Check each repository’s license and dependencies: for example, change-matcher distinguishes its MIT code from its Apache-2.0 base model. The homepage also links GabbleGrid and Zibajoon as separate products, plus an enterprise lab still in stealth. Their availability and terms should be checked at their destinations. No portfolio-wide data-retention or model-training policy was found.
Frequently asked questions
How do I start using a project?
Follow its repository link and requirements. citation-verifier offers the documented Python installation, tests and contrasting value-verification examples above; other projects have different environments.
Are the public components complete applications?
Not necessarily. Several READMEs explicitly identify a published component and describe what remains private. Review that boundary before planning an integration.
Does open source include free model usage?
No model-service allowance comes with reading the portfolio. One Roof uses your agent accounts, and optional hosted models require your own API key. Assess code licensing and service costs separately.
Official code and further reading
- Gaurav Narasimhan on GitHub: repositories, README files and project-specific issue links.
- Engineering blog: the author’s explanations of agent workflows and engineering decisions.
Alternatives to Gaurav Narasimhan


One Roof
Bring coding-agent sessions, approvals and bounded loops into one Mac app.
AI Coding & Developer ToolsFree

Sendoff
Draft and send newsletters from Claude or Codex on your own cloud accounts.
AI Marketing & Sales ToolsPaid

UnoRouter
One key for multiple AI models, with provider routing and browser chat
AI Model APIs & InfrastructureFreemium

Wavefunction
Parallel coding tasks with your own cloud or local models.
AI Coding & Developer ToolsFree

kwant
Run your own AI team with handoffs, schedules and approvals.
AI Automation ToolsFree

Infino
Open-source retrieval over Parquet, combining keywords, vectors and SQL for AI applications.
AI Databases & Context InfrastructureFreemium
- Pricing
- Free
- Website
- gaurav8936.github.io
- Added
- November 2026
- Updated
- November 16, 2026
More in Tool & Resource Directories
Add the badgeShow you're listed on AITroveX
<a href="https://aitrovex.com/tool/gaurav-narasimhan?ref=badge" target="_blank">
<img src="https://aitrovex.com/badge-listed-light.svg" alt="Listed on AITroveX" style="height: 54px; width: auto;" />
</a>