TaskPilot AI

Your Personal AI Chief of Staff — Conquering Engineer Task Overload with Autonomous Multi-Agent Intelligence

Built for the DELL FutureMind AI Hackathon by Team IdeaForg-E

📑 Complete Documentation Suite

TaskPilot AI is accompanied by dedicated technical documentation:

Document	Focus & Scope	Link
🏛️ System Architecture	Multi-layer topology, dual-tier LLM engine, SQLite WAL persistence, design trade-offs	docs/architecture.md
⚡ Pipeline Deep-Dive	End-to-end data transformation stages, mathematical formulas, P1 injection	docs/pipeline.md
🤖 Agent Internals	Deep technical specifications for all 8 agents, heuristics, validation, prompts	docs/agents.md
🔌 API Reference	Complete REST contract, Pydantic schemas, request/response examples	docs/api.md
📊 Interactive Diagrams	Standalone explorable architecture and user journey diagrams via Archify	Archify Visualizations

The Problem
Our Solution
System Architecture
Interactive Architecture Diagrams (Archify)
Multi-Agent Pipeline
Tech Stack
Getting Started
Environment Variables
Project Structure
API Overview
Demo Walkthrough
Team IdeaForg-E
License

The Problem

Modern software engineers are drowning in context fragmentation. Critical work items arrive across disparate, disconnected tools: issue boards, Slack threads, customer email escalations, calendar invites, and meeting transcripts. Without a single pane of glass, engineers suffer severe cognitive overload:

Operational Friction	Impact on Engineering Productivity
🔀 Source Fragmentation	Developers juggle 4 to 7 tools daily to track action items
⏳ Context Switching Tax	Every interruption requires an average of 23 minutes to regain focus
👻 Invisible Task Debt	~35% of actionable obligations buried in emails and chat are untracked
🎯 Priority Blindness	Engineers optimize locally for the loudest alert, not global impact
📋 Triage Overhead	45+ minutes lost each morning manually summarizing notes and emails

Our Solution

TaskPilot AI is an autonomous multi-agent assistant that acts as a personal chief of staff for software engineers:

📥 Autonomously aggregates tasks across 5 heterogeneous data sources (GitHub, Slack, Email, Calendar, Meetings).
🔍 Extracts hidden action items from unstructured conversational text using targeted few-shot LLM reasoning and regex heuristics.
🔗 Deduplicates and correlates cross-platform duplicate signals into unified Master Tasks via string and token similarity.
🛡️ Audits quality and completeness across 7 dimensions, automatically drafting clarification questions for vague tickets.
⚖️ Intelligently prioritizes using a 7-factor weighted scoring model with explainable rationale and anti-noise safeguards.
📅 Compiles dynamic daily plans that lock fixed calendar commitments and auto-inject decompression breaks to prevent burnout.
💬 Enables interactive copilot dialogue with natural language retrieval and autonomous mid-day P1 incident re-prioritization.

Before vs After TaskPilot

Workflow Stage	Traditional Developer Routine	With TaskPilot AI
Morning Routine	Open 5+ browser tabs, manually read threads	Open TaskPilot: see a single, ranked, calendar-aware plan
Hidden Obligations	~35% forgotten in Slack threads or email chains	Extracted automatically with calibrated confidence scoring
Prioritization	Subjective gut-feel or loudest stakeholder	Objective 7-factor algorithmic scoring + explainable tags
Mid-Day Emergency	Manually re-read schedule, panic reschedule	Chat: "Inject P1" $\rightarrow$ incremental re-rank in seconds
Daily Time Saved	Baseline	2+ hours saved per engineer daily

System Architecture

TaskPilot AI uses a cooperative multi-agent architecture where 8 specialized agents collaborate through a shared SQLite database in WAL mode.

flowchart TB
    subgraph DS["Heterogeneous Data Sources"]
        direction LR
        GH["GitHub Issues & PRs"]
        SL["Slack Messages & Threads"]
        EM["Email Inboxes & Alerts"]
        CA["Calendar Schedule"]
        MT["Meeting Transcripts"]
    end

    subgraph ORCH["Agent 0 — Orchestrator"]
        OC["Pipeline Controller & State Machine\n(Stale Run Recovery & Diagnostic Guard)"]
    end

    subgraph PIPELINE["Multi-Agent Processing Pipeline"]
        direction LR
        A1["Agent 1\nIngestion"]
        A2["Agent 2\nExtraction"]
        A3["Agent 3\nFusion"]
        A4["Agent 4\nQuality"]
        A5["Agent 5\nPriority"]
        A6["Agent 6\nPlanner"]
        A1 --> A2 --> A3 --> A4 --> A5 --> A6
    end

    subgraph CHAT["Agent 7 — Interactive Copilot"]
        C1["Natural Language Q&A\nAutonomous P1 Incident Injection"]
    end

    subgraph DB["Shared State Persistence (SQLite WAL)"]
        SQL[("SourceEvents | TaskCandidates\nMasterTasks | ContextLinks\nQualityReports | PriorityScores\nDailyPlans | TimeSlots")]
    end

    subgraph FE["React 18 Dashboard"]
        direction LR
        P1["Dashboard"]
        P2["Tasks"]
        P3["Quality"]
        P4["Priority"]
        P5["Planner"]
        P6["Chat"]
    end

    DS --> ORCH --> PIPELINE
    PIPELINE <--> DB
    CHAT <--> DB
    CHAT -.->|Triggers Incremental Run| ORCH
    DB <--> FE

Note

For a comprehensive breakdown of architectural tiers, entity relationships, and engineering trade-offs, read System Architecture Blueprint.

Interactive Architecture Diagrams (Archify)

Explore interactive system diagrams generated with Archify:

Diagram	Technical Focus	Link
🤖 Agent Architecture	Multi-agent execution pipeline with 8 specialized stages	Open Interactive Diagram
🔄 User Flow Journey	End-to-end user navigation from ingestion to daily schedule	Open Interactive Diagram
🏗️ Application Architecture	Full-stack topology connecting React, FastAPI, Groq, and SQLite	Open Interactive Diagram

Features: Search nodes, trace connected dependencies, switch light/dark themes, and export high-res PNG/SVG artifacts.

Multi-Agent Pipeline

The core pipeline transforms raw signals through 6 sequential stages plus an interactive copilot:

Stage	Agent Name	Primary Execution Logic	LLM Model	Offline Fallback
0	Orchestrator	Coordinates sequential execution & tracks state	None	Sequential execution loop
1	Ingestion	Parses 5 raw JSON feeds into uniform `SourceEvent` records	None	Native JSON parsers
2	Extraction	Discovers explicit tickets & hidden tasks with confidence scoring	`openai/gpt-oss-20b`	Regex action heuristics
3	Fusion	Correlates duplicate items via string & token similarity	None	`difflib.SequenceMatcher`
4	Quality	Audits completeness across 7 dimensions (0-100 rubric)	`openai/gpt-oss-20b` (critical)	Rule-based QA rubric
5	Prioritization	Calculates 7-factor weighted score & anti-noise penalties	`openai/gpt-oss-20b` (batch of 8)	7-factor weighted formula
6	Planning	Builds calendar-aware schedule with decompression breaks	`openai/gpt-oss-120b`	Greedy slot allocator
7	Chat Copilot	Answers schedule queries & triggers incremental P1 injection	`openai/gpt-oss-20b`	Template message responses

Tip

Total Pipeline Speed: Full pipeline completes in ~14-18 seconds on Groq Cloud LPUs, and in < 2 seconds in zero-token local fallback mode (TASKPILOT_DISABLE_LLM=1).

Learn more in the Pipeline Deep-Dive and Agent Internals.

Tech Stack

Backend & AI Engine

Runtime: Python 3.11+
API Framework: FastAPI with Pydantic v2 data validation
ORM & Database: SQLAlchemy 2.0 + SQLite with WAL mode (PRAGMA journal_mode=WAL)
LLM Hardware: Groq Cloud Language Processing Units (LPU)
LLM Models:
- openai/gpt-oss-20b: Ultra-fast extraction, quality auditing, batch prioritization, and chat (~1.0s latency).
- openai/gpt-oss-120b: High-capability multi-constraint schedule optimization (~2.5s latency).
Resilience: Circuit breaker (2 consecutive failures $\rightarrow$ 60s cooldown), rate-limit backoff, and truncated JSON auto-repair.

Frontend Dashboard

Framework: React 18 SPA built with Vite 5
Styling: Tailwind CSS v4 (responsive dark/light themes)
Routing: React Router v6
Client: Axios with diagnostic and error hint interceptors
Components: Lucide React icons, React Markdown renderer

Getting Started

Prerequisites

Python 3.11+
Node.js 18+ (with npm)
A free Groq API Key from console.groq.com

Quick Start (Windows)

:: 1. Clone repository
git clone https://github.com/IdeaForg-e/TaskPilot-AI.git
cd TaskPilot-AI

:: 2. Configure environment
copy backend\.env.example backend\.env
:: Edit backend\.env and insert your GROQ_API_KEY

:: 3. Launch full stack with one click
start.bat

start.bat automatically launches the FastAPI backend on http://localhost:8000 and the React frontend on http://localhost:5173.

Manual Setup

Backend Setup (FastAPI)

cd backend
python -m venv venv

# Activate virtual environment
# Windows:
venv\Scripts\activate
# Linux/macOS:
# source venv/bin/activate

pip install -r requirements.txt
cp .env.example .env
# Edit .env and configure GROQ_API_KEY=gsk_your_key

uvicorn app.main:app --reload --port 8000

API docs available at: http://localhost:8000/docs

Frontend Setup (React + Vite)

cd frontend
npm install
npm run dev

Dashboard available at: http://localhost:5173

Environment Variables

Configure backend/.env:

# ==============================================================================
# Primary LLM Provider (Groq Cloud LPU)
# ==============================================================================
GROQ_API_KEY=gsk_your_actual_groq_api_key_here

# Model Overrides (Optional)
GROQ_MODEL_FAST=openai/gpt-oss-20b
GROQ_MODEL_REASONING=openai/gpt-oss-120b

# ==============================================================================
# Database Persistence
# ==============================================================================
DATABASE_URL=sqlite:///./taskpilot.db

# ==============================================================================
# Development Flags
# ==============================================================================
# Set to 1 to run entirely in zero-token local fallback mode (no LLM calls)
TASKPILOT_DISABLE_LLM=0

# Clean database tables on startup (default: false to preserve data)
CLEAN_DB_ON_STARTUP=false

Project Structure

TaskPilot-AI/
├── backend/
│   ├── app/                       # FastAPI core application
│   │   ├── main.py                # Server entry point & global error handler
│   │   ├── database.py            # SQLite connection & WAL pragma configuration
│   │   ├── models/                # 8 SQLAlchemy relational entities
│   │   ├── routers/               # 9 modular API route files (routers 0 through 8)
│   │   ├── schemas/               # Pydantic request/response schemas
│   │   └── services/              # Business logic controllers (Agents 0 through 6)
│   ├── agents/                    # AI Agent core implementations
│   │   ├── llm_client.py          # Resilient Groq client, circuit breaker, JSON parser
│   │   ├── agent_2_extraction_agent.py
│   │   ├── agent_2_validation.py  # 6-step candidate validation filter
│   │   ├── agent_3_fusion_agent.py
│   │   ├── agent_4_quality_agent.py
│   │   ├── agent_5_prioritization_agent.py
│   │   ├── agent_6_planning_agent.py
│   │   └── prompts/               # Versioned prompt templates
│   ├── tests/
│   │   ├── golden_dataset.json    # Accuracy test dataset (11 scenarios)
│   │   └── test_accuracy.py       # Precision, recall, and F1 validation suite
│   ├── requirements.txt
│   └── .env.example
├── frontend/
│   ├── src/
│   │   ├── pages/                 # Dashboard, Tasks, Quality, Priority, Planner, Chat
│   │   ├── components/            # Steppers, modal dialogs, schedule timelines
│   │   ├── services/api.js        # Axios client with error unwrapping
│   │   └── context/               # Theme & session context
│   └── package.json
├── data/                          # 5 input JSON feeds (~50 developer records)
├── docs/                          # Technical documentation & Archify diagrams
│   ├── architecture.md            # System architecture blueprint
│   ├── pipeline.md                # Data transformation pipeline deep-dive
│   ├── agents.md                  # Comprehensive agent specifications
│   ├── api.md                     # Complete REST API reference
│   ├── agent-architecture.html    # Interactive agent diagram
│   ├── application-architecture.html # Interactive full-stack diagram
│   └── user-flow.html             # Interactive user journey diagram
├── start.bat                      # One-click Windows launcher
├── README.md
└── LICENSE

API Overview

All routes are served under /api/v1. Full schemas and response payloads are in docs/api.md.

Domain	Method	Endpoint	Description
Orchestrator	`POST`	`/orchestrate/run`	Triggers asynchronous 6-stage pipeline
	`GET`	`/orchestrate/status/{run_id}`	Polls real-time stage progress & status
	`GET`	`/orchestrate/latest`	Fetches latest run metrics, latency & accuracy
Ingestion	`POST`	`/ingest`	Parses raw JSON files into `SourceEvent` rows
Extraction	`POST`	`/extract`	Extracts explicit tickets & hidden tasks
Fusion	`POST`	`/fuse`	Correlates and merges duplicate candidates
Quality	`POST`	`/quality/evaluate`	Evaluates task completeness across 7 dimensions
	`GET`	`/quality/reports`	Retrieves all task quality audit reports
Prioritization	`POST`	`/prioritize`	Computes 7-factor scores & assigns rank order
	`GET`	`/tasks/ranked`	Fetches the ranked priority Leaderboard
Tasks	`GET`	`/tasks`	Lists all MasterTasks with status/source filters
	`GET`	`/tasks/{task_id}`	Returns 360° telemetry: task, quality, priority & context links
	`POST`	`/tasks/{task_id}/status`	Updates task status (`open`, `in_progress`, `done`)
Planning	`POST`	`/daily-plan`	Compiles calendar-aware schedule with buffer hours
	`GET`	`/daily-plan/{date}`	Retrieves daily plan & timeslot schedule
	`GET`	`/planner/calendar`	Groups tasks with deadlines by date
Chat & P1	`POST`	`/chat`	Conversational copilot & autonomous P1 injection
System	`GET`	`/health`	Liveness probe & LLM configuration status

Demo Walkthrough

Follow this 6-step walkthrough to experience TaskPilot AI's full feature set:

flowchart LR
    Step1["1. Dashboard\nRun Pipeline"] --> Step2["2. Tasks\nInspect Sources"]
    Step2 --> Step3["3. Quality\nAudit Completeness"]
    Step3 --> Step4["4. Priority\nLeaderboard"]
    Step4 --> Step5["5. Planner\nCalendar Schedule"]
    Step5 --> Step6["6. Chat Copilot\nInject P1 Emergency"]
    Step6 -.->|Auto Re-rank| Step1

Dashboard: Navigate to the Dashboard and click "Run Pipeline". Watch the live 6-stage stepper execute across Ingestion, Extraction, Fusion, Quality, Prioritization, and Planning in ~15 seconds.
Tasks Explorer: Open the Tasks view to explore the unified master backlog. Use source badges (GitHub, Slack, Email, etc.) to filter, and click any item to inspect its provenance.
Quality Auditor: Visit the Quality page to view the circular QA gauge. Review which tasks are classified as Actionable vs Needs Info, and inspect AI-generated clarification questions.
Priority Leaderboard: Check the Priority tab to see tasks mathematically ranked from #1 to #N. Click any card to view the 7-factor breakdown and explainable narrative.
Calendar Planner: Open the Planner to see the synthesized workday. Notice how fixed meetings from calendar.json are protected and decompression coffee breaks are automatically inserted.
Chat Copilot & P1 Injection: Open the Chat tab and type:

"inject a P1 defect - payment gateway down in production"

Watch the system extract incident details, modify the source datastore, trigger an incremental pipeline re-run, and announce that the new emergency has captured the #1 rank on your schedule.

Team IdeaForg-E

Built for the DELL FutureMind AI Hackathon by:

Team Member	Engineering Responsibility
Disha	Backend Lead (FastAPI architecture, SQLite database, API routers), Calendar Scheduler Agent, Copilot chat
Priyanka	Agent Engineer 1 (Ingestion normalizer, Extraction agent, Fusion deduplicator)
Chaitanya	Agent Engineer 2 (7-dimension Quality auditor, Prioritization formula)
Disha + Jagruti	Frontend Engineers (React 18 SPA, Tailwind CSS v4 dashboard, responsive UI)
Anil	Integration Lead (Orchestrator, pipeline lifecycle, LLM resilience, deployment)

License

This project is licensed under the MIT License — see the LICENSE file for details.

Built with care by Team IdeaForg-E for the DELL FutureMind AI Hackathon

Navigation Menu

Repository navigation

README.md

README.md

TaskPilot AI

📑 Complete Documentation Suite

Table of Contents

The Problem

Our Solution

Before vs After TaskPilot

System Architecture

Interactive Architecture Diagrams (Archify)

Multi-Agent Pipeline

Tech Stack

Backend & AI Engine

Frontend Dashboard

Getting Started

Prerequisites

Quick Start (Windows)

Manual Setup

Environment Variables

Project Structure

API Overview

Demo Walkthrough

Team IdeaForg-E

License

FilesExpand file tree

README.md

Latest commit

History

README.md

File metadata and controls

TaskPilot AI

📑 Complete Documentation Suite

Table of Contents

The Problem

Our Solution

Before vs After TaskPilot

System Architecture

Interactive Architecture Diagrams (Archify)

Multi-Agent Pipeline

Tech Stack

Backend & AI Engine

Frontend Dashboard

Getting Started

Prerequisites

Quick Start (Windows)

Manual Setup

Environment Variables

Project Structure

API Overview

Demo Walkthrough

Team IdeaForg-E

License