An engineering blog-style deep dive into building an AI-powered sales productivity platform, currently in active development for its upcoming public launch.
ClosePitch was conceived with a clear vision: to empower individual sales professionals and lean startup founders to close more business with less administrative overhead. The long-term product direction concentrates on transforming manual sales coordination pipelines into automated, intelligent workflows that deliver customized communications, track telemetry, and optimize conversion efficiency.
Modern sales professionals waste up to 40% of their day writing follow-up emails, managing lead pipelines, and logging interactions. Traditional CRMs are complex, slow on mobile devices, and require significant manual data entry. Furthermore, generic outreach copy converts poorly; sales teams need customized pitch messaging tailored to each lead's unique business parameters.
Existing customer relationship tools fail on three major fronts:
To solve these problems, we made several strategic product scoping choices:
We prioritized a mobile-first, high-performance interface designed for quick interaction:
All core actions (adding leads, choosing AI prompt templates) operate inside sliding bottom drawers, keeping context visible without full-screen navigation loads.
Adding a lead or updating a status modifies the UI immediately, while caching writes to SQLite and carrying out sync requests in the background.
Every technology in the ClosePitch stack was chosen based on clear performance and maintainability metrics:
Allowed building pixel-perfect custom interfaces for iOS and Android using a single Dart codebase. Reusable layout structures halved our time-to-market.
Go microservices compile to native binary, offering sub-millisecond execution times, native concurrency, and small memory foot prints under load.
Excellent relational consistency, transaction isolation, and support for JSONB data formats to store unstructured lead parameters alongside tables.
Ensures environment consistency across local development, CI/CD testing pipelines, and AWS production container deployment.
Click on any node in the system diagram below to explore the technical implementation details of that specific layer.
Cross-platform mobile app running on iOS and Android. Implements BLoC state management and feature-first clean architecture directory structures.
The frontend code utilizes a feature-first clean architecture pattern:
Core logic is split into decoupled microservices communicating via an API Gateway. Built using standard Go packages and lightweight libraries to avoid bloat:
Rather than direct API pipelines, ClosePitch uses dynamic context layers to optimize AI model outputs:
Retrieves user profile, company details, target lead's website summary, and selected pitch objective.
Fuses context into a structured engineering system prompt. Demands output formatting in valid JSON schema.
Orchestrated via Go backend API calls to Gemini API. Enforces schema constraints and safety configs.
ClosePitch implements a hybrid data approach inside PostgreSQL:
Services compile into lightweight Docker containers deployed to AWS infrastructure:
To keep mobile response times fast, we implemented several performance configurations:
During development, we resolved several critical technical hurdles:
Large Language Models frequently returned malformed JSON or markdown syntax inside JSON blocks, breaking parsing components.
Solution: Implemented system-prompt-level schema enforcement combined with validation checks in Go, auto-retrying with adjusted temperature configs on failure.
Sales professionals lost connection inside subways or airports, leading to lost pitch drafts.
Solution: Engineered local SQLite transaction layers that auto-save drafts locally, queueing them up for sync once connections restore.
ClosePitch has delivered clear outcome improvements based on production tracking metrics:
Key retrospectives from design, build, and deployment:
Interact with the roadmap below to see the upcoming feature priorities for the ClosePitch product: