.tv-fallback{font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;padding:2rem;color:#86efac;background:#030712;line-height:1.6}.tv-fallback a{color:#4ade80}.tv-fallback h1,.tv-fallback h2,.tv-fallback h3{color:#bbf7d0}.tv-fallback ul{padding-left:1.25rem}.tv-fallback li{margin-bottom:.5rem}

tvsguide.io · Projects · Two Bear Capital API

Two Bear Capital API

Three-phase AI extraction API for VC pitch decks with Claude Sonnet 4.5

Two Bear Capital API is a multi-service API with async job-based PDF processing using ARQ workers. Built with FastAPI, PostgreSQL, Redis, and designed for Railway deployment.

The V3 API implements a CRUD-style three-phase workflow: upload PDF (sync), convert to images (async), and extract data (async). This architecture enables granular failure tracking, efficient retries of only failed phases, and resource reuse for multiple extractions per PDF.

AI extraction uses Claude Sonnet 4.5 to analyze pitch decks and extract comprehensive data including company info, team, traction, market analysis, and fundraising details. Includes life sciences support for indication, modality, and development stage. Features evidence-based citations with page numbers and verbatim quotes, anti-hallucination safeguards (returns null for missing data), and per-section confidence scoring.

Offers both real-time extractions and batch processing with 50% cost savings. Includes webhook support for job completion notifications.

Tech: Python, FastAPI, Redis, PostgreSQL, Claude AI, ARQ, Railway, Supabase

Features

Three-phase extraction workflow: Upload → Convert → Extract
AI-powered pitch deck analysis with Claude Sonnet 4.5
Evidence-based citations with page numbers and confidence scoring
Batch extractions with 50% cost savings (up to 24h latency)
Life sciences support: indication, modality, development stage
Webhook notifications for job completion
Multi-layer PDF validation with hyperlink sanitization

More from the portfolio

This project is part of tvsguide.io, the personal portfolio of Tim Veil — software engineer and CIO at Two Bear Capital, previously at StarTree, Cockroach Labs, and Hortonworks. The full collection covers distributed systems, data infrastructure, JDBC drivers, AI services, build pipelines, real-time analytics, and a couple of personal apps. Each project ships with source code, tech notes, and links to live deployments where applicable.

BigQuery JDBC Driver — Modern JDBC driver for Google BigQuery, optimized for database IDEs with fast metadata support
Flight Chop — Passenger focused platform helping travelers understand turbulence and other conditions that may impact their flight
Leaves of Blocks — Free SwiftUI iOS puzzle game with no ads, no tracking, and no in-app purchases
Bloviate — Java library for generating realistic test data for databases
Quantum Beacon — Pipeline for analyzing company domains for Post-Quantum Cryptography (PQC) support
BenchBase — Multi-DBMS SQL benchmarking framework via JDBC
Two Bear Tracker — Portfolio tracking application with comprehensive analytics monitoring capabilities
Two Bear Capital MCP Server — Model Context Protocol server for enterprise data access
Bear Stack — Modern monorepo template with clean architecture patterns
Fivetran Affinity Connector — Custom Fivetran connector for Affinity CRM data sync
Fivetran Carta Connector — Custom Fivetran connector for Carta equity management
Fivetran Standard Metrics Connector — Custom Fivetran connector for portfolio company metrics
Fivetran dbt — dbt project transforming multi-source data into BigQuery
Pinot TestContainers Example — Integration testing framework for Apache Pinot using TestContainers
CockroachDB Remote Client — Docker container for one-time CockroachDB administrative tasks
CockroachDB Single Node — CI/CD optimized Docker image for CockroachDB testing
CockroachDB Dynamic Certs — Dynamic certificate generator for secure CockroachDB deployments
CockroachDB Dynamic HAProxy — Dynamic HAProxy load balancer for CockroachDB clusters
CockroachDB Spring Examples — Spring Boot examples for connecting to CockroachDB

Browse the project index, read about Tim's background, follow the blog, or return to the homepage. Source code lives on GitHub; professional history is on LinkedIn.