AI & Context Engineering Projects

Agentic deliveryWorking prototype

Relay

From task context to production, with nothing skipped.

Why it matters Agents move fast; Relay keeps ownership, evidence and sign-off auditable for every change.

tests: 105
stages: 15
human gates: 4

Fifteen stages and four human approval gates
Commit-bound evidence; approvals reset when their basis changes
Local dashboard and offline demo, standard-library Python only

Python
SQLite
JavaScript
CLI adapters

Scope: Single-user local prototype. Live GitHub, Jira and production delivery are not verified end to end.

github.com/sidd105/agentic-feature-delivery-workflow

PAY-2C84 · Add a health endpoint to the payments service

Context
Plan
Build & PR
Review
QA
Deploy

Blocked · PR reviewReview found blocking issues (grade F). The developer pushes a fix and the review runs again on the new head.

human approval gate · a new commit resets tests, review and QA

Try it locally

git clone https://github.com/sidd105/agentic-feature-delivery-workflow.gitcd agentic-feature-delivery-workflowpython3 -m sdlc_orchestrator.dashboard

Context engineeringEarly compiler

Context Engineering SDLC

The right organizational knowledge for the task.

Why it matters Agents and engineers get the rules that apply to this task, not the whole wiki.

units in the sample org: 62
knowledge layers: 5
tests: 37

Five knowledge layers with ownership, scope and must/should/may levels
Relevance scoring, overrides, conflicts and gap reporting
CLI, library and MCP server over one deterministic engine

TypeScript
YAML
MCP

Scope: Deterministic, primarily lexical matching. Compiled context still needs inspection before use.

github.com/sidd105/context-engineering-sdlc

$ ctx explain "Add a new payment provider"

Business4/10
Domain12/18
Architecture7/10
Code8/12
Operations7/12

38 of 62 units · token budget1,499 / 1,500

selected for this task left out, with reasons

Try it locally

git clone https://github.com/sidd105/context-engineering-sdlc.gitcd context-engineering-sdlcnpm cinpm run ctx -- explain "Add a new payment provider"

Code qualityLocal tooling

Context-Aware PR Reviewer

Code review grounded in your organization’s rules.

Why it matters Every pull request is held to the same org standard, and every deduction is explainable.

tests: 80
evidence connectors: 5
report formats: 3

Org-locked policy that a pull request can’t weaken
AI findings kept separate from deterministic evidence
Skipped checks reported as coverage gaps, never as passes

TypeScript
Python AST
SARIF
MCP

Scope: Reads working-tree files; commit-bound CI assessments require a clean checkout of that commit.

github.com/sidd105/context-aware-pr-reviewer

payments-service

40.3/100

SecurityF
Code qualityF
Org standardsF
MaintainabilityD
TestingB-
ArchitectureB

Gate critical-security-caps-at-C · ACME-SEC-002 −45 (Snyk, critical)

Try it locally

git clone https://github.com/sidd105/context-aware-pr-reviewer.gitcd context-aware-pr-reviewernpm cinpm run example

Distributed systemsLocal demo

Distributed AI Rate Limiter

One shared budget. Across every API instance.

Why it matters Scaled-out services stop overspending a shared AI quota.

concurrent requests: 1,000
admitted, exactly: 100
Node processes: 4

Atomic multi-bucket admission in a single Redis Lua script
Token reservations, refunds and excess-usage debt
Concurrency tests across four Node processes, plus k6 load scenarios

TypeScript
Node.js
Redis
Lua
k6

Scope: Simulated inference; production authentication and settlement recovery remain outside the demo.

github.com/sidd105/distributed-ai-rate-limiter

api-1
api-2
api-3

Shared Redis bucket

100 capacity

EVALSHA admit.lua

100 admitted HTTP 200
900 rejected HTTP 429

Try it locally

git clone https://github.com/sidd105/distributed-ai-rate-limiter.gitcd distributed-ai-rate-limiterdocker compose up --build

What I’m building

The Sovereign Ship Loop

The tools behind it

Relay

Context Engineering SDLC

Context-Aware PR Reviewer

Distributed AI Rate Limiter