Skip to content

Submission: Cognost — self-improving stakeholder-alignment Company Brain - #38

Open
smutyala1at wants to merge 3 commits into
topoteretes:mainfrom
smutyala1at:submission/cognost
Open

Submission: Cognost — self-improving stakeholder-alignment Company Brain#38
smutyala1at wants to merge 3 commits into
topoteretes:mainfrom
smutyala1at:submission/cognost

Conversation

@smutyala1at

Copy link
Copy Markdown

Every product team has the same failure mode: the PRD says one thing, the roadmap
drifted, Sales signed a contradicting promise, and the design shipped a feature
marked "Won't have." A normal RAG bot papers over this — it picks one source and
sounds confident. Cognost does the opposite.

It surfaces where the docs disagree, cites both sides, and says which value is
current vs. superseded — then improves its own answering behaviour from feedback.

What we built

  • Ingest — 42 BrainFlow documents (PRD, ADRs, roadmap, decision log, Slack,
    transcripts, metrics CSV) into a Cognee Cloud knowledge graph with typed edges:
    contradicts · supersedes · owned_by · cites
  • Query — GRAPH_COMPLETION recall with a versioned wiki-maintainer skill;
    every answer cites its source and names any conflict
  • Lint — catches contradictions, superseded-but-still-referenced decisions,
    metric drift, and ownership orphans — reported, never auto-resolved
  • Self-improve — scores each answer, records failures as session feedback,
    proposes a skill rewrite sourced from the CLAUDE.md maintainer contract, applies explicitly

smutyala1at and others added 3 commits June 19, 2026 20:44
Team: Maria Beiner, Rozy Kassab, Santosh Mutyala, Martin Kaiser
Repo: https://github.com/kaiser-data/cognost

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…ence files

- SUBMISSION.md: team details, slides, 5-point pitch, skills table, before/after evidence
- my_skills/wiki-maintainer/SKILL.md: improved skill with provenance + contradiction rules
- my_skills/wiki-maintainer/SKILL.baseline.md: baseline v1 for before/after comparison
- evidence/before-after.md: scored before/after proof (baseline 1/10 → maintainer 10/10)
- evidence/lint-report.md: all 11 planted misalignments caught
- evidence/demo-cases.md: 3-minute demo cue sheet with verified live cases

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant