Kora turns a company's raw pitch deck or financial documents into a structured, evidence-backed due-diligence profile — automatically extracted, automatically checked for internal inconsistencies, and always traceable back to the original source text.
It is an independently designed and built product: backend, data model, AI pipeline, and frontend design system are all original work, built end-to-end by one person.
Early-stage due diligence is still mostly manual. An analyst opens a pitch deck, reads it, re-types the numbers into a spreadsheet, cross-checks whether the growth story actually adds up, and separately keeps a mental list of what the founder still hasn't disclosed. That process is slow, inconsistent between analysts, and easy to get wrong — a plausible-sounding deck is not the same as an internally consistent one, and it's easy to miss that "12-month runway" and "$1M/month burn" and "$6M cash" don't actually agree with each other.
Kora automates the mechanical part of that process — extraction, consistency checking, gap-finding — so a human analyst spends their time on judgment instead of data entry, while every number Kora shows stays traceable back to the sentence it came from.
Kora is built for the analyst side of early-stage investing and corporate development: VC/angel analysts, individual investors, and deal teams who need to triage a stack of pitch decks and financial documents and quickly decide which ones deserve a deeper look — without a large associate team to do the first pass for them.
Document → Extract → Validate → Analyze → Ask → Report
- Document — an analyst uploads a pitch deck, financial statement, or data-room export (PDF, DOCX, or TXT) into an organization (a workspace shared with their team).
- Extract — an LLM extracts structured financial facts (revenue, ARR, MRR, burn rate, runway, CAC, LTV, and more) and qualitative facts (business model, market, team, risks), each one tied back to the page/passage it was extracted from via a source citation — not just a number floating with no provenance.
- Validate — a separate, deterministic (non-LLM) rules engine checks the extracted facts against each other: does the stated LTV:CAC ratio match what the underlying numbers actually compute to? Is a growth claim backed by more than one data point? Is a "profitable" claim contradicted by the reported net income? These checks are plain Python business logic, unit-tested, and never delegated to a model that could talk itself into a wrong answer.
- Analyze — Kora assesses evidence coverage against a due-diligence checklist (company, financial, market, team), computes a category-weighted investment score from the verified facts — and deliberately withholds that score, showing no number instead of a misleading one, when coverage is too thin or critical fields are missing.
- Ask — once a document is indexed, the analyst can ask an AI chat assistant follow-up questions grounded in the source documents (retrieval-augmented generation), with every answer citing the passages it drew from — including an "analytical" mode that can query the extracted financial data directly instead of only paraphrasing text.
- Report — the full analysis — extracted facts, findings, coverage, score, and cited chat context — can be exported as a due-diligence report (Markdown or PDF), and every analyzed document rolls up into a per-organization portfolio view.
- AI-driven extraction of financial and qualitative facts, each with a source citation back to the original document
- Deterministic validation findings (inconsistency checks) that are always distinguished in the UI from the LLM's own document-stated facts and inferences
- A due-diligence coverage checklist that explicitly shows what's missing and why it matters, not just what was found
- An investment score that can be null by design — Kora never fabricates a number on thin evidence
- Retrieval-augmented chat with citations, plus a tool-calling mode that computes answers from real extracted data
- Exportable due-diligence reports (Markdown / PDF), with two report formats (a narrative report and an evidence-grounded report with verified facts, red flags, and founder questions)
- Multi-organization workspaces with role-based membership and portfolio-level analytics across every analyzed company
Backend — Python 3.12, FastAPI, SQLAlchemy 2.0 (async) + Alembic, PostgreSQL with the pgvector extension for embeddings, JWT auth (pwdlib/argon2 password hashing), LLM calls (extraction, chat, embeddings) routed through OpenRouter's OpenAI-compatible API, pypdf/reportlab for document parsing and PDF report generation. Tested with pytest (203 tests across services and integration).
Frontend — Next.js 15 (App Router) with React 19 and TypeScript, TanStack Query for server state, a custom dark-themed "Kora" component library (Radix UI primitives underneath), Tailwind CSS.
Infrastructure — Docker Compose (PostgreSQL/pgvector + FastAPI backend), local filesystem document storage behind a storage-backend interface designed for a future S3-compatible swap.
See docs/architecture.md for the full system diagram and a walkthrough of how the pieces fit together, and docs/product-case-study.md for the product reasoning behind the platform, including honest trade-offs and current limitations.
kora-platform/
├── backend/ FastAPI + PostgreSQL (pgvector) API — see backend/README.md
├── frontend/ Next.js 15 app (the Kora UI) — see frontend/README.md
├── docs/ Architecture, product, and UI documentation
├── docker/ Container assets for deployment
└── docker-compose.production.yml
# Backend — see backend/README.md for full setup
cd backend
uv sync
cp .env.example .env # then fill in SECRET_KEY, POSTGRES_PASSWORD, OPENROUTER_API_KEY
uv run alembic upgrade head
uv run fastapi dev app/main.py --port 8000
# Frontend — see frontend/README.md for full setup
cd frontend
npm install
npm run devThe frontend runs at http://localhost:3000, the backend at http://localhost:8000 (interactive API docs at /docs). See backend/README.md for database setup (a local pgvector/pgvector Docker container is the fastest path) and required environment variables.
docs/product-case-study.md— the problem, the users, the product decisions, and honest trade-offs/limitationsdocs/architecture.md— system diagram, workflow diagram, and a short walkthrough of each layerdocs/portfolio-description.md— CV/portfolio/interview-ready descriptions of this projectbackend/README.md— backend setup, configuration, running locally, API overviewfrontend/README.md— frontend setup, environment, scriptsdocs/ui/design-system.md— the Kora visual design system (tokens, primitives, usage)docs/company/anddocs/prd/— earlier product/business planning documents from an initial concept exploration that predates the current implementation; kept for history, not a spec of the shipped product
See LICENSE.