Use ML to Predict Delivery Time and Improve Code Reliability
summary
Built a predictive pipeline that extracts features from Git history, runs RAG + LLM reasoning, and outputs time estimates with confidence, risk scores and recommendations; demo uses Python FastAPI, Chroma, OpenRouter/Cloud API, analyzes a hackathon project split into backend (Node.js + Postgres), frontend (Koko game engine) and smart contract. Estimates lift on-time delivery from ~60% to ~80%, flags commits/PRs with quality <50% or risk >50%, catches critical security issues, and reports ~0.1% false positives; demo repo is not production ready.
AI-written summary · about
chapters (11)
- 0:00Intro: AI-Driven Engineering Quality
- 1:13The Solution: AI Pipeline for Delivery-Time Prediction
- 2:24How It Works: Feature Extraction + RAG + LLM Reasoning
- 3:10Demo Setup & Hackathon Repo Overview
- 3:57Repo Risk Breakdown: Backend vs Frontend Complexity
- 4:44Discovery-Time Estimates: PMs Get Instant Timelines
- 5:52Part 2 — AI Code Review on Commits & PRs
- 7:44Security Horror Show: “Bad Code” Example Analysis
- 8:25Comparing AI Estimates vs Actual Delivery Time
- 9:15Wrap-Up: What We Built + Results + Tech Stack
- 11:01Get the Demo on GitHub + Final Thanks
related talks
26mDelivering Delivery Time Prediction
24mUncovering Community and Project Insights Through Data Driven Methods
16mPractically predictable – Forecasting delivery when nothing is certai
17mR & Python playing nice, in Production
44mAgile metrics for predicting the future
25mSupercharging Pipeline Efficiency with ML Performance Prediction
50mML Deployment Methodology Illustrated by Route Time Estimation
43mPredictable Delivery