founder, thothica new delhi · updated 25 jul 2026

Adnan Abbasi

Most of what I make is a library in disguise.

An archive, a publisher's backlist, a stream of regulations: I model them until people and AI agents can search them and cite them properly. The trade is called data ontology. I mostly just call it getting the shelves right.

company
Thothica, an AI-native knowledge systems company in New Delhi, incorporated 2023. Studio Whence, its imprint, turns those pipelines into finished books and comics.
open source
Falsafa, a library of world philosophy that any AI agent can call over MCP. There is no vector database anywhere in it. The librarian looks things up and quotes its sources.
method
Structure, connect, operationalize: model the entities, map the relations between them, then give agents a way to act on the result.

the live register

7 entries

Every link below answered over HTTPS on 25 July 2026. This list stays to what is live and reachable right now, the way anyone else would find it, not what is planned or sitting in staging.

  1. thothica.com company live · 25 jul 2026

    The company. What Thothica builds, for whom, and why the work keeps coming back to ontology.

  2. falsafa.ai open source live · 25 jul 2026

    A living library of philosophy and classics: 2,018 works and around 1.375 million citable passages, readable by people and callable by agents. Funded by an Emergent Ventures grant.

  3. indianliberals.in client · ccs live · 25 jul 2026

    The Centre for Civil Society's archive of Indian liberal thought, rebuilt from a slow WordPress site into a fast static library: 1,018 primary works, 725 thinker pages, an authority file of 454 thinkers.

  4. mcp.indianliberals.in client · ccs live · 25 jul 2026

    The archive's MCP server. Point an AI agent at it and the agent can search the collection and quote it, citations attached.

  5. ocrwithai.com product live · 25 jul 2026

    Thothica OCR. Upload a scanned book; it measures the type on every page, recovers the real chapter structure, and returns a clean reflowable edition.

  6. docs.thothica.com docs live · 25 jul 2026

    Platform documentation, public.

  7. meowbench.com side project live · 25 jul 2026

    Which AI draws the best cat. 29 models, 6 prompts, 4 attempts each, 3 vision judges, and a crowd Elo arena. Rigorous about something unserious.

thothica ledger

a product · demonstrators

Revenue intelligence for state governments: read the registers a state already keeps, link them into one graph, rank what deserves a second look. Three editions, built on Thothica's own initiative on synthetic data. None was commissioned by a state government, and none is a customer.

client engagements

named with permission

The same method each time, pointed at very different problems. A few engagements stay unnamed.

  • Diamond Pocket Books publisher · comics + research

    One of India's largest mass-market publishers. Research dossiers distilled from books and interviews, then comic scripts generated against them, with 5,214 line-level source citations behind a verify gate so no panel ships an unsourced fact. The first four books are in print.

  • Pareeksa universities · on-screen exam marking

    An on-screen marking and assessment platform for universities and Olympiads, anonymous by construction. The marking app never holds a student's identity, no name, no roll number, no PII. De-anonymization lives in a separate reconciliation app across a storage boundary, so a grader can never see whose paper it is.

  • Handy marketplace · home services

    A two-sided marketplace: booking, bidding, payments and payouts on append-only ledgers, ratings, parts logging. Runs on Cloudflare D1 and R2. Demo deployed.

  • CricToday media · editorial pipeline

    An editorial pipeline that catches AI-written drafts before they publish: burstiness scoring, persona ban flags, staged rewrites. English and Hindi. In verification.

  • Thothica back office internal · in daily use

    Our own GST invoicing and cash flow tool, with signed invoices and an MCP server, so an AI client can create and look up invoices by itself.

open source

  • Falsafa flagship

    The library above. The whole corpus is flat, citable files served from R2, on purpose: an agent gets a librarian, never a similarity score.

  • MinerU True Copy agpl · desktop

    A desktop OCR app built on Tauri. A fork of OpenDataLab's MinerU that I maintain, with a vision model OCR sidecar and an Indic translation path. github.com/adoistic/mineru-true-copy

  • Naql translation atlas

    An open atlas of how books crossed languages: isnad-graded transmission chains, a typeset book, an MCP server over the result. github.com/adoistic/naql

  • pdf2fxl document tooling

    Turns image-only picture-book PDFs into fixed-layout EPUB3 and editable PPTX: OCR for text and layout, inpainting to erase the baked-in text, then deterministic fit-to-box typesetting. github.com/adoistic/pdf2fxl

  • pdf-toolkit document tooling

    PDF tools: auto-generated tables of contents, and high-fidelity conversion to Word. github.com/adoistic/pdf-toolkit

  • forme template

    An editorial magazine CMS for print. One operator, offline. github.com/adoistic/forme

  • founder-office-hours claude skill

    A published Claude skill: the office-hours method for a non-technical founder paired with a technical co-founder, building an AI-native startup. github.com/adoistic/founder-office-hours

  • personal-svelte-template template

    An accessibility-first personal website template: SvelteKit and Sveltia CMS, fork it and edit through a plain admin screen. github.com/adoistic/personal-svelte-template

  • studio-whence-press print tooling

    A client-side RGB-to-CMYK print converter (FOGRA52). Nothing leaves the browser. github.com/adoistic/studio-whence-press

recognition

  • 2026 Winner, MSME Idea Hackathon 5.0. Ministry of MSME, Government of India. ₹12.75 lakh, one of 502 ideas selected nationwide.
  • 2026 "The Prayer the Machine Cannot Pray", Philosophy Now, Issue 173. On AI and occasionalism.
  • 2025 Emergent Ventures grant. Tyler Cowen's program at the Mercatus Center. $20,000, for Falsafa.
  • 2024 Cobden-Bright Award. For an AI persona of F. A. Hayek, later used by the Friedrich Naumann Foundation in a podcast.

writing & research

Philosophy Now ran my essay on AI and occasionalism in Issue 173: whether a machine can pray, and why I think today's language models cannot. A longer essay on language models and the public sphere is in draft.

I also write closer to the machine. One recent study takes an LLM pipeline that curates technical diagrams from arXiv and asks whether a small model on a laptop can replace the paid LLM gates. A frozen SigLIP2 probe classifies diagrams at 86 percent, more than ten points above the LLM that generated its own training labels, at about 52 milliseconds a crop.

Before any of this I published poetry. It still shapes how I edit.

about

I came to software through the humanities. Literature and philosophy first, then a college international relations course where I convinced the professor to let me do AI text analysis instead of a term paper. That project ended up at MPSA in Chicago, and I never did write the term paper. No CS degree; a humanities one, on a full merit scholarship, with four years on the Dean's List.

The through line is Hayek's argument that progress comes from aggregating dispersed knowledge. Every system on this page moves knowledge across a barrier of language, format, or access. I hold to occasionalism, in the line of al-Ghazali and Malebranche. It is a minority position. I am fine with that.

New Delhi. My father is my co-founder.