PeakTalk

AI-assisted stress testing for arguments before difficult work meetings.
AI-стресс-тест аргументов перед сложными рабочими встречами.

PeakTalk helps a professional test a concrete position—not generic confidence— before a roadmap review, budget defense, stakeholder challenge, or similar high-stakes conversation. A user supplies working material, examines its weak points, answers adversarial questions in a text-first simulation, and leaves with a compact Defense Brief.

Project status: test-stage engineering snapshot. This public repository is a one-time, sanitized view of the product source and its development history. It is suitable for code review and local evaluation, but it is not a public production deployment or an operational runbook.

Adversarial simulation screen with synthetic content. No customer, interview, or production data is shown.

What the product does

Working material
      ↓
Weak-point and pressure scan
      ↓
Text-based adversarial simulation
      ↓
Defense Brief: arguments, evidence gaps, risks, and next steps

The core contract is deliberately narrow:

• input is a specific initiative, roadmap, budget, memo, or set of arguments;
• observations should be tied to the supplied material or marked as assumptions;
• the simulation tests likely objections instead of scoring charisma or diction;
• typing is always supported; voice input is an optional way to compose text;
• the final artifact is meant to be useful immediately before the real meeting.

See the core-flow specification and MVP boundaries for the observable product scope.

System overview

flowchart LR
    U["Browser"]
    FE["Next.js application<br/>UI, server routes, authentication"]
    API["FastAPI service<br/>product and admin APIs"]
    DB[("PostgreSQL<br/>product and auth data")]
    REDIS[("Redis<br/>cache and task broker")]
    WORKER["Celery worker<br/>document and background jobs"]
    OBJECTS[("Private object storage<br/>source documents")]
    AI["LLM provider adapter"]
    EXT["Email, payment, and push adapters"]

    U --> FE
    U --> API
    FE --> DB
    API --> DB
    API --> REDIS
    REDIS --> WORKER
    API --> OBJECTS
    WORKER --> OBJECTS
    API --> AI
    WORKER --> AI
    FE --> EXT
    API --> EXT
    WORKER --> EXT

Loading

The diagram describes component responsibilities, not a production network or deployment topology. The public snapshot intentionally contains no production access material.

Area	Technology	Role
Web	Next.js 16, React 19, TypeScript	Product UI, server-side auth routes, responsive interaction
API	FastAPI, Pydantic, SQLAlchemy	Product contracts, authorization, orchestration
Persistence	PostgreSQL, Alembic	Auth and product records, schema evolution
Background work	Celery, Redis	Parsing, scheduled jobs, cache and task transport
Document handling	S3-compatible storage, PDF/DOC/DOCX/TXT/MD parsers	Private source material and text extraction
AI integration	OpenAI-compatible provider adapter	Weak-point analysis, adversarial prompts, final brief
Quality	Pytest, Node test runner, ESLint, TypeScript, CodeQL	Contract, build, and security checks

More detail is available in Architecture.

Local evaluation

The quickest isolated check does not require external AI, payment, email, or storage credentials:

cd backend
python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.lock
pytest

cd ../frontend
npm ci
npm run lint
npm run typecheck
npm test

For the local application stack, copy the safe examples and start Compose:

test -f backend/.env || cp backend/.env.example backend/.env
test -f frontend/.env.local || cp frontend/.env.example frontend/.env.local
docker compose up --build

The examples contain localhost values or explicit placeholders only. Do not replace them with production credentials. Provider-dependent flows remain unavailable until intentionally configured in a separate local environment.

See Local development for ports, migrations, manual startup, and the full verification commands.

Repository map

backend/                 FastAPI application, workers, migrations, tests
frontend/                Next.js application and frontend contract tests
docs/
  architecture.md        Components, boundaries, and data flow
  local-development.md   Reproducible local setup and checks
  product/               Product scope and non-goals
  specs/                 Observable behavior
  decisions/             Selected public-safe engineering decisions
  project-history.md     How to inspect the preserved Git history
  research-insights.md   Anonymized, derived discovery insights
.github/                 Review-safe CI, security checks, and governance
THIRD_PARTY_NOTICES.md   Redistributed assets and their upstream licenses

History and provenance

This is not a new repository reconstructed as a single synthetic commit. Its source history contains 351 rewritten historical commits, including 18 merge commits, before the current review-hardening commit. Original author and committer timestamps, ordered parent relationships, merge topology, commit subjects, and five selected branch references were preserved. Commit IDs changed because sensitive content and attribution metadata were filtered; historical cryptographic signatures therefore cannot be carried over as valid.

Some retained commits are content-empty after excluded material was removed. They remain in the graph so that the sequence and merge structure are not silently rewritten into a cleaner-looking story.

Read Provenance for the exact public accounting and Project history for useful Git commands. The private old-to-new commit map is intentionally not published.

AI coding assistants were used during parts of development and review. They are tools, not Git authors; human ownership and review responsibility remain with the project maintainer. This is disclosed in more detail in Provenance.

Security

• Real secrets are never valid example configuration.
• Production deployment workflows, access instructions, and infrastructure topology are outside this repository.
• CI uses least-privilege permissions and does not deploy production systems.
• Please do not publish suspected vulnerabilities, credentials, or user material in an Issue or pull request.

Use Security policy to report a vulnerability privately.

Known limitations

• The product remains in test stage; this snapshot is not a claim of production readiness, availability, or security certification.
• Discovery evidence is a small, non-representative sample. Derived insights are hypotheses to validate, not market proof.
• AI output may be incomplete or wrong and must be reviewed against the source material before a real decision.
• Full document, email, payment, push, and AI flows require external services; local placeholder configuration deliberately does not impersonate them.
• This is a fixed review snapshot and is not continuously synchronized with the private production repository.

Documentation

Document	Purpose
Architecture	Component responsibilities and trust boundaries
Local development	Setup, migrations, and checks
Core flow	Observable preparation workflow
MVP boundaries	In-scope and explicitly excluded outcomes
Project history	Preserved graph, selected branches, and caveats
Research insights	Anonymized functional implications from discovery
Provenance	Sanitization and authorship disclosure
Security	Private vulnerability reporting
Contributing	Review workflow and engineering standards
Third-party notices	Redistributed assets and upstream license terms

Rights

The source is publicly visible for engineering review, but it is not released under an open-source license. See NOTICE. No additional license is granted beyond the GitHub Terms of Service, applicable law, or a separate written agreement.