Build Your Own Research Assistant — Projects, Notebooks and RAG

In this session

1. 01

   The opening mirror: a junior's prompt about a live matter goes up (names, amounts, a diagnosis-shaped fact) and the room votes on whether it is safe to send. The identifiers are highlighted and the abstraction move performed live: the same question about 'a licensee' and 'a sum' gets the same quality of answer with nobody's client in it. Behind the rule: consumer free tiers log conversations and most, by default, train on them. The opt-out switches exist and the lab flips them. A workspace makes the stakes permanent, because what you upload persists.

2. 02

   RAG, drawn once and properly: your documents are stored and indexed; at each question a retriever fetches the passages that look most relevant; those go into the window; the model answers from them, ideally with a pointer to where. Every strength follows: the corpus can be far bigger than any window, answers come with addresses, and the model is reshaping rather than recalling.

3. 03

   The three ways the clerk fails: the retrieval miss (the right passage never fetched, so the answer rests on the wrong ones), the wrong pointer (a citation to a passage that says something adjacent), and the override (the model answering from its training anyway, especially when the sources contradict what it 'remembers'). Grounded is better; grounded is not true: the Stanford benchmark of professional legal-research tools found even citator-backed systems hallucinating on roughly one query in six.

4. 04

   The harness zoo, demystified with one question for all four artefacts: what persists? Custom instructions: a standing preamble to every chat (free on the major tiers; small; the right home for 'always cite paragraph numbers; flag anything unverified'). A Project (ChatGPT: free tier, five files per project; Claude: free tier, five projects): instructions plus files plus chats in one container, a small RAG over your bundle. A Gem (Gemini, free): a saved instruction set with optional knowledge files. NotebookLM (free; 50 sources per notebook, each up to half a million words): the purest: source-grounded by design, every answer carrying inline citations that click through, and, unusually for a free tool, not training on uploads.

5. 05

   Two doors. One closed: building a custom GPT is no longer a free-tier exercise (at the time of checking, the vendor's own pages disagreed about which paid plans still allow it); the free-tier route is Projects, Gems and notebooks, and it is enough. One opened: ChatGPT Projects can now be shared, even on free accounts, so a moot team can work one workspace together, with the caveat that a shared project trains on members' data unless every member has turned training off.

6. 06

   The instruction set is standing orders written like office policy, where a law student's training shows: answer only from the sources in this workspace; cite document and paragraph for every proposition; where the sources do not address the question, say 'not in the sources' rather than supplement from general knowledge; where sources conflict, say which is later and which is higher; keep the research log when asked. Each line answers something Session 2 taught. This is harness engineering, policy rather than code: instructions for a junior who is brilliant, literal, and never says 'I don't know' unless ordered to.

7. 07

   The live build, on the bundle the course has been assembling: State v. Navjot Sandhu, Anvar P.V. v. P.K. Basheer, Shafhi Mohammad, Arjun Panditrao Khotkar and s.63 of the Bharatiya Sakshya Adhiniyam: the electronic-evidence line, at once a doctrine-tracing exercise (watch the certificate requirement fall, rise, waver and settle across four cases, with pinpoints) and the law Session 7 will need. Then the trap question the bundle cannot answer, and the workspace, under its instruction set, says 'not in the sources', the most professionally reassuring sentence a machine can produce.

8. 08

   The staleness catch, the moment the session is for. The workspace is asked who may sign the s.63(4) expert certificate, and answers from its newest source as if the law stopped there, because for this workspace it did. Then Pune Bar Association v. Union of India (SC, 22 May 2026), which upheld the certificate regime and read 'expert' beyond the notified s.79A examiners, is added to the bundle live; the question is re-asked; the answer changes, with a new pinpoint. A workspace is a snapshot, not a subscription: it knows nothing it was not given, including last month's law. Updating the bundle is now part of updating the matter; 'is my bundle current?' joins 'is it still good law?' on the standing checklist.

9. 09

   What never goes in, now as doctrine rather than caution: a consumer workspace is a standing disclosure to a third party, so client documents, instructions and identifying facts stay out, full stop. The legal frame, previewed here and completed in Session 7: privilege under s.132 BSA belongs to the client and is not yours to waive into a training corpus; the professional rules make the client's confidence your standing duty.

10. 10

    The statutory frame arriving on top: the DPDP Act (substantive obligations commencing May 2027) will make the advocate a data fiduciary engaging an unbriefed processor the moment client data is pasted; and In Re: Summoning Advocates (2025 INSC 1275) has already shown the Supreme Court thinking hard about what a shared repository of a lawyer's files exposes. The working rule the course will not soften: published law in, clients never. Enterprise-grade exceptions are something firms buy; free tiers do not give them.

11. 11

    When not to build one: a workspace earns its keep on recurring work over a stable bundle: a moot, a seminar paper, an internship's running matter, exam revision. For a one-off question, Session 3's paste-and-ask beats it. The diagnostic vocabulary grows one word: alongside the prompt ceiling and the model ceiling there is a tool ceiling. Last session's two-hundred-page judgment, which defeated the free chat window, fits inside one free NotebookLM source. Sometimes the answer is a better-shaped tool rather than a better model, and it is free.

The build-along

The room builds it and keeps it.

The class builds the thing with the presenter, the mechanism is explained as it is built, and a named skill closes the hour.