Skip to content

TODO.md

Latest commit

 

History

History
133 lines (94 loc) · 5.02 KB

TODO.md

File metadata and controls

133 lines (94 loc) · 5.02 KB

TODO for typeagent-py

Organizing principle

Tasks are grouped by component/feature area, with priorities P0-P3 (P0=higest, P3=lowest). Note that priorities were assigned by Copilot, and I don't always agree (it often assigns P0 or P1 to complicated work items).

Each item also has a rough size estimate.

Meta-TODO

Gradually move work items from here to repo Issues.

Additional code reviews (of AI-vibe-coded things)

P1 - High Priority

  • [P1, medium] Scrutinize sqlite/reltermsindex.py
  • [P1, medium] Review the new storage code more carefully, adding notes

Storage & Persistence

P1 - High Priority

  • [P1, large] Unify tests for storage APIs
  • [P1, medium] Make (de)serialize methods async in interfaces.py if they might execute SQL statements
  • [P1, medium] Implement SqliteRelatedTermsAliases.serialize() and other missing serialize()/deserialize() methods

P2 - Medium Priority

  • [P2, large] Fix all bugs related to ordinals/ids relying on starting at 0 and no gaps (prepare for deletions)
  • [P2, large] Refactor memory and sqlite indexes to share more code (e.g. population and query logic)
  • [P2, medium] Make the collection/index accessors in StorageProvider synchronous (the async work is all done in create())
  • [P2, medium] Replace the storage accessors with readonly @property functions
  • [P2, medium] "Ordinals" should be renamed to "Id" (tedious though)

P3 - Low Priority

  • [P3, medium] Flatten secondary indexes into Conversation (they are no longer optional), reducing the structure to:
    • Message collection
    • SemanticRef collection
    • SemanticRef index
    • Property to SemanticRef index
    • Timestamp to TextRange
    • Terms to related terms
  • [P3, medium] Split related terms index in two (aliases and fuzzy_index)
  • [P3, large] Implement consistent approach to deletions (tombstoning in sqlite, cascade delete semrefs and indexes)

Query & Search Pipeline

P1 - High Priority

  • [P1, medium] Look more into why the search query schema is so unstable
  • [P1, large] Redesign the whole pipeline; make each stage its own function with simpler API
  • [P1, large] Improve test coverage for search, searchlang, query, sqlite

P2 - Medium Priority

  • [P2, medium] Implement token budgets for answer generation (may leave out messages, favoring only knowledge)
  • [P2, medium] Change answer context to be text (message texts and timestamps), not JSON or semantic ref ordinals
  • [P2, medium] Split large answer contexts to avoid overflowing the answer generator's context buffer

Type System & Interfaces

P1 - High Priority

  • [P1, medium] Move TypedDicts out of interfaces.py (they don't belong there)
  • [P1, medium] Fix need for # type: ignore comments (22 in typeagent/) by making I-interfaces more generic

P2 - Medium Priority

  • [P2, medium] Sort out why IConversation needs two generic parameters
  • [P2, medium] Simplify TTermToSemanticRefIndex generic parameter
  • [P2, medium] Tighten types: several places allow None and construct default instances; either disallow None or skip that functionality

Serialization

P2 - Medium Priority

  • [P2, medium] Remove a bunch of XxxData TypedDicts that can be dealt with using deserialize_object and serialize_object
  • [P2, small] Catch and report DeserializationError better
  • [P2, medium] Look into whether Pydantic can do our (de)serialization (presumably faster?)

Code Quality & Cleanup

P1 - High Priority

  • [P1, large] Make coding style more uniform (e.g. docstrings)
  • [P1, large] Reduce code size

P2 - Medium Priority

  • [P2, small] Avoid most inline imports
  • [P2, medium] Break cycles by moving things to their own file if necessary
  • [P2, medium] Unify or align or refactor VectorBase and EmbeddingIndex
  • [P2, medium] Address TODO comments (too numerous)
  • [P2, medium] Address raise NotImplementedError("TODO") (five found) -- implement it

P3 - Low Priority

  • [P3, medium] Change inconsistent module names (Claude uses different naming style)
  • [P3, medium] Rewrite podcast parsing without regexes
  • [P3, medium] Switch from Protocol to ABC
  • [P3, medium] Reduce duplication between ingest_vtt.py and typeagent/transcripts/
  • [P3, small] Rename kplib.py to something ending in _schema.py

Testing

P1 - High Priority

  • [P1, large] Add new tests for newly added classes/methods/functions

P2 - Medium Priority

  • [P2, medium] Review Copilot-generated tests for sanity and minimal mocking

Documentation

P1 - High Priority

  • [P1, small] Document test/build/release process
  • [P1, small] Document how to run evaluations (but don't share the data)

P2 - Medium Priority

  • [P2, large] Document low-level APIs (key parts used directly by high-level APIs, e.g. ConversationSettings)

Development Infrastructure

P3 - Low Priority

  • [P3, small] Move typeagent into src/
  • [P3, tiny] Move test/ to tests/