Skip to main content

ENGINEERING KAM

A research engine
that can show its work.

Kam turns fragmented sports-market observations into qualified research surfaces. The architecture separates evidence, computation, agent reasoning, and presentation so a fluent answer can never quietly become the source of truth.

research.request
01 observe.source_event02 preserve.s3_evidence03 qualify.rust_worker04 publish.dynamodb_view05 reason.agentcore06 render.workspace_receipt

REQUEST LIFECYCLE

From observation
to inspectable answer.

Every stage owns one job and emits an explicit contract. That keeps collection running independently from developer inference and lets the product replay, repair, or replace a projection without rewriting history.

01

Observe

Connectors collect source observations as events. Each observation keeps its source, event time, and identity instead of becoming an anonymous number.

02

Preserve

Immutable evidence lands in S3. Reprocessing starts from durable facts, not from a model answer or a mutable screen state.

03

Qualify

Rust services validate identity, freshness, coverage, and publication rules. Events that cannot support a claim do not silently enter the read model.

04

Project

Purpose-built projections shape the latest useful view for each product access pattern: board, game, workspace, memory, and receipt.

05

Research

AgentCore coordinates bounded tools over qualified views. Chat can navigate the workspace, but it cannot rewrite the underlying evidence.

06

Publish

The answer and screen ship with freshness, coverage, exclusions, and lineage so the user can inspect what supports the conclusion.

SYSTEM MAP

Durable on the left.
Responsive on the right.

INPUTSource eventsodds · scores · markets · user actions
EVIDENCEAmazon S3immutable objects · replayable lineage
COMPUTERust servicesvalidation · identity · qualification
READDynamoDBaccess-pattern projections
AGENTAgentCorebounded tools · context · orchestration
PRODUCTKam workspacechat · screens · receipts · saved work

DESIGN INVARIANTS

The rules that keep
research honest.

Evidence before language

The model explains qualified facts; its prose never becomes the system of record.

Events before shared state

Producers publish observations. Consumers build their own bounded projections without coupling every workflow together.

API responses match screens

The future Rust Web API will return purpose-built product views instead of exposing storage details to the client.

Receipts travel with answers

Freshness, missing sources, qualification state, and lineage stay visible at the point of use.

ARCHITECTURE STATUS

The native product is the checkpoint. The backend is next.

Kam’s public website is static and makes no research-data or storage calls. The successor backend is being designed around a versioned Rust Web API, AgentCore, and an event-driven pipeline. This page describes that direction without presenting unfinished work as deployed fact.

See how the workspace fits together.

Explore the static interaction model, then open Kam to try the local chat preview.

Explore the workspace Return to product