Noesis
A governance layer for companies putting AI agents into production. Every action monitored, every decision auditable, and a human in the loop wherever the stakes require one. Agents get reviewed the way a team does.
The place where experiment is always brewing
Things I am building to find out whether they work.
Side projects, prototypes, and interface experiments. Some are infrastructure I intend to turn into a company. Some exist because I wanted to see what would happen. Each has a short write-up with the architecture and the current state, including the parts that are not finished.
Outr
Researched outbound for founder-led B2B sales. Finds companies matching your ideal customer, gathers real evidence they are ready to hear from you, and drafts a specific email. Nothing is sent until you approve it.
Agent evaluation platform
Tracing and evaluation for enterprise AI agents. Records what an agent actually did, scores it with automated checks and human review, and catches regressions before a release reaches production.
unslop-text
Paste in a draft and get an audit of the tells that make prose read as machine-written. A regex pass runs entirely in the browser, then an optional model pass catches the structural problems no pattern can see.
Ask me anything
A voice agent that answers questions about my work in my own voice. Tap the microphone, speak, and pause. It answers out loud and links through to the longer version of whatever it just told you.
The machine-readable profile
A personal site whose home page is addressed to AI agents rather than recruiters. Serves structured data, debate prompts written to resist flattery, and an adversarial mode that invites a visitor's model to audit the career record for inconsistencies.
Time travel interface
One career rendered in four complete design systems: a 1998 CRT terminal, 2012 skeuomorphism, a 2024 glass dashboard, and a 2035 archive. The facts stay fixed and only the visual language and vocabulary change.
Book writing agents
A two-agent loop that rewrites long-form text into a target style and length. A RewriteAgent drafts each section, a CheckerAgent scores it for human-likeness and rule adherence, and a Noesis policy gate blocks banned patterns deterministically before the checker ever runs.
Aura
A personal AI clone. Clone your voice from audio samples, load a private knowledge base, then ask it questions and hear the answers back in your own voice. Retrieval runs as in-process BM25, no vector database or embedding key required.
Nanda demo
A working prototype of the NANDA agent-discovery protocol: a client resolves an agent by name, follows signed pointers through a two-hop index-to-facts lookup, verifies everything with real Ed25519 cryptography, and rejects a single tampered field.
Knowledge graph
An interactive force-directed graph of my career, ventures, kReative Labs projects, and writing, and how they actually connect. Every node and edge traces back to something already public on this site.
More to come. If something here is relevant to what you are working on, I am happy to talk about it: shripathmit@gmail.com