TinyTroupe

repository·main·Indexed 27 days ago

https://github.com/microsoft/tinytroupe

A framework for LLM-based people simulation used for design, validation, and insight generation in business. TinyTroupe enables the orchestration of multiple AI agents (TinyPerson) to perform complex tasks through collaborative interaction. It includes tools for managing experiments via InPlaceExperimentRunner, extracting structured data with ResultsExtractor, and steering agent behavior using the Intervention class. The framework supports OpenAI models and provides experimental support for Ollama.

Tokens
22.6K
Snippets
24
Records
100
Agent score
92%

What's inside tinytroupe

  1. Understand TinyTroupe's intended use and limitations

    main

    TinyTroupe is an experimental Python library designed for simulating human behavior using TinyPerson agents in TinyWorld environments. It is a tool for research and analysis, not an AI assistant for direct user interaction.

    Intended Uses

    • Analysis of artificial human behavior through simulation.
    • Generation of synthetic artifacts.
    • Supplementing human insight generation.
    • Researching computational cognitive architectures.

    Non-Intended Uses (Warnings)

    • DO NOT use TinyTroupe for direct interaction with end-users. Developers must implement their own Responsible AI layer for any user-facing products.
    • DO NOT use simulations for policy or consequential decision-making. Simulation results may not reflect reality and can have real-world impacts if misused.
    • DO NOT allow simulations to control real-world mechanisms without strict damage control mechanisms in place.
  2. Execute community engagement and mobilization

    main

    Use the following scripts and materials to inform communities and reduce misinformation:

    Community Engagement Script (Field Use):

    • Greet and introduce the team/purpose.
    • Explain vaccines offered, eligibility, and safety information.
    • Explain the observation period and symptoms to report (dizziness, breathing difficulty, swelling).
    • Invite questions and provide follow-up contact.

    CHW Door-to-Door Mobilization Script:

    "Hello, my name is [CHW name] with [organization]. On [date], we will be offering [vaccine name] at [site/time]. If you are [eligibility], please come between [hours]. We will observe you for 15 minutes after vaccination. If you can't attend, we will visit your area on [date]. For any concerns contact [CHW phone]."

    Minimum IEC Materials Required:

    • One-page flyer: Include vaccine type, eligibility, location/time, what to bring, and what to expect.
    • Site entrance poster: Include a flow diagram and safety basics.
  3. Follow the internal approval workflow for communications

    main

    To ensure legal and community compliance, all external communications must follow this four-step internal approval workflow:

    1. Drafting: Lisa prepares the draft message and posts it to the ops channel.
    2. Legal Review: Legal counsel reviews for legal risk and Indigenous land-claim implications (Target timeline: 12 hours).
    3. Community Review: Marcos reviews for cultural and community engagement implications (Target timeline: 12 hours).
    4. Final Sign-off: Wild Advice leads (Lisa and Oscar) provide final sign-off. External sending is permitted only after step 3 is complete.
  4. Data collection requirements for BlueHarbor scoping

    main

    To produce credible scenario modelling and investor-grade outputs within the 6-week window, the following datasets must be collected and uploaded by the primary owners (Alex for technical/data and Lisa for stakeholder liaison):

    A. Operational & Network Data (Owner: Alex)

    • Load profiles: 12–24 months (15-min preferred; hourly acceptable). Provide separate files per feeder/class.
    • Outage/event logs: Time-stamped for the same period, including tags for planned vs unplanned and cause codes.
    • Asset registers: Gensets (make, model, rated efficiency curves), transformers, critical switchgear, meters, and SCADA logs.
    • Protection & Diagrams: Current protection settings and one-line diagrams.
    • GIS Data: Network/GIS maps & shapefiles for candidate sites and feeders (with declared coordinate system).

    Requested Filename Convention:

    • islandname_load_15m_YYYYMM.csv
    • islandname_outage_log_YYYYMM.csv

    B. Fuel & Financial Data (Owner: Alex / Lisa)

    • Diesel consumption history and recent fuel contracts (supply terms, indexation, minimum volumes).
    • Tariff schedules (by customer class), recent collections, accounts receivable aging, and cross-subsidy notes.
    • Historical monthly fuel pricing (delivered price) and hedging/forward contract details.
    • Lender/donor pre-conditions, term-sheet drafts, or reporting templates.

    C. Technical & Site Constraints (Owner: Oscar)

    • Site access: Landowner agreements, security, crane/road limits, and night work constraints.
    • Permitting: Timelines for civil works and electrical interconnection.
    • Environmental/Social: Shoreline setbacks and cultural sites.

    D. Procurement & Vendors (Owner: Sara)

    • Preferred battery/inverter/containerised system vendors and regional lead-times.
    • For each vendor: Typical lead time (weeks) for a 500 kW/1 MWh system, warranty terms, local spares availability, and after-sales support.
    • Existing vendor contracts, procurement templates, and policy constraints (e.g., single source, local content).

    E. Stakeholder Contacts (Owner: Lisa)

    • Primary client contacts: CEO/MD, Technical Director, CFO/Treasury, Head of Procurement, regulator liaison.
    • Donor/lender or regulatory points-of-contact (email, phone, preferred contact hours).
    • Suggested meeting windows for Day-1 kickoff and weekly touchpoints.
  5. Data checklist for rapid scoping (24–48h priorities)

    main

    To initiate the rapid scoping process, the following datasets must be secured within 24–48 hours. Owners are assigned to ensure accountability:

    • Technical Data (Owner: Alex):
      • 12–24 months of load profiles (15-min preferred)
      • Outage/event logs (time-stamped)
      • Diesel consumption history & recent fuel contracts
      • Genset specifications & asset registers
      • Tariff schedules & customer segmentation
    • Technical/Siting Data (Owner: Oscar):
      • Network/GIS shapefiles & candidate site notes
    • Procurement Data (Owner: Sara):
      • Vendor contacts & lead-times (batteries, inverters, containers)
    • Stakeholder Data (Owner: Lisa):
      • Lender/donor pre-conditions or term-sheet templates
      • Primary client contacts (CEO/MD, Technical Director, CFO/Treasury, Procurement, regulator)
  6. Execute the 0–24h Immediate Action Plan

    main

    The first 24 hours focus on demonstrating engagement to the regulator, securing data access, and establishing communication channels.

    Action Items:

    • Submit Status Memo: Upload a 1-page regulator-ready status memo to BlueHarbor/ZephyrPay/AML/Deliverables/Status_Memo_Lisa.md to confirm engagement and outline the high-level plan.
    • Secure Contacts: Obtain the ZephyrPay Point-of-Contact (POC) and legal counsel details, and secure a copy of the original regulator notice.
    • Setup Secure Data Transfer: Create a destination folder on a secure platform (SFTP, secure cloud share, or secure enclave) and document the URL, permitted user list, and transfer protocol.
    • Request Data Sample: Request a pseudonymised sample extract of $\le$1000 rows that matches the exact schema. If hashing is not possible, request secure enclave access or a synthetic dataset.
    • Legal Readiness: Ensure legal counsel is on standby to review regulator wording and customer communication scripts, providing redlines within 4 hours of a draft.
    • Establish Cadence: Set up a daily 09:00 stand-up calendar invite for the regulator POC and ZephyrPay leadership.
    # Sample 24h status memo template
    - Brief greeting and summary of receipt of notice.
    - Statement of engagement and named Wild Advice Partners lead.
    - High-level plan and timeline (72h triage + 3-week remediation).
    - Ask list and deadlines (sample, secure folder, POCs).
    - Request for regulator POC for daily updates and confirmation of preferred delivery format.
  7. Configure the secure enclave for data analysis

    main

    If client-side hashing is not permitted, use a secure enclave. The environment must meet the following requirements:

    • Isolation: An isolated environment provided by ZephyrPay or a third party.
    • Access Control: Controlled user accounts and mandatory session recording.
    • Data Restrictions: No copy/paste or data export allowed except for pre-approved aggregate outputs.
    • Support: Live remote session support for Wild Advice Partners.
    • Scheduling: A 2–4 hour enclave session should be scheduled within D1–D3 for rapid analytics.
  8. Execute procurement and FX mitigation

    main

    Follow these principles for equipment and supplier procurement:

    • Localization: Prioritize local/regional suppliers to minimize customs, FX risk, and mobilization delays.
    • Contingency: Prepare conditional hold/standby options for long-lead items (e.g., drill rigs, specialized pumps).
    • Currency: Quote in local currency where possible. If FX exposure is unavoidable, use client-led FX mechanisms for rapid payments.
    • Compliance: Ensure suppliers provide appropriate insurance and safety documentation.

    Equipment Specifications

    • Toe buttress rockfill: Crushed or imported quarry rock, nominal D50 = 100–300 mm, well-graded angular material.
    • Geotextile: Non-woven separation fabric (tensile strength per engineer spec).
    • Sandbags: Polypropylene (UV-stabilized).
    • Silt curtain: Floating polypropylene with weighted skirt.
    • Pumps: Diesel-driven submersible pumps (preliminary assumption: 100–300 L/s cumulative flow capacity).
  9. Execute immediate procurement and logistics plan

    main

    Prioritize the following procurement actions within the first 48 hours:

    Immediate Priorities:

    • Vaccine carriers with validated temp-loggers (including spares).
    • Passive cold boxes for fixed sites.
    • PCM packs (sufficient for outreach windows).
    • Temp logger spare units.
    • Basic PPE, syringes, sharps boxes, and waste bags.
    • Lightweight tents or local shade shelter alternatives.

    Procurement Path:

    1. Check existing GreenRiver-approved suppliers first.
    2. Use local suppliers for tents, fuel, and small items to reduce lead time.
    3. For imports (solar fridges, loggers), use emergency procurement channels and check for duty waivers.

    Logistics Notes:

    • Consolidate pick-up points; assign a dedicated owner for last-mile logistics.
    • Pre-condition PCM packs and check carrier temperatures before loading.
  10. Execute the MultiLever Rapid Diagnostic Plan

    main

    The Rapid Diagnostic is a 5-business-day process designed to identify primary drivers of demand decline (price elasticity, income effects, channel shifts, etc.) and produce SKU/category-level elasticity estimates.

    Objective: Produce a 1-page executive diagnostic highlighting the top 10 demand drivers and top 20 SKUs by lost sales potential, alongside a prioritized list of micro-tests (price, pack-size, or targeted promotion).

    Roles & Responsibilities:

    • Lisa: Rapid synthesis, lead for diagnostic write-up, and stakeholder communication.
    • Alex: Data ingestion, validation, and provisional elasticity model outputs.

    Timeline:

    • Day 0: Kickoff and data requests.
    • Day 1–2: Data ingestion, cleaning, and exploratory analysis.
    • Day 2–4: Rapid modelling and synthesis.
    • Day 5: Delivery of 1-page diagnostic, data appendix, and client briefing.
    | Rank | SKU | Category | QoQ Demand Δ (%) | Estimated Price Elasticity | Confidence (High/Med/Low) | Immediate Action |
    |---:|---|---|---:|---:|---|---|
    | 1 | SKU-12345 | Food – Staple rice 2kg | -28% | -1.6 | Med | Trial smaller 1kg value-pack at -5% price |
    | 2 | SKU-23456 | Home appliance – Toaster | -32% | -2.1 | Low | Targeted promotion to loyalty cohort |
  11. BlueHarbor 6-week analytical project phases

    main

    The BlueHarbor scoping project follows a 4-phase analytical approach over 6 weeks:

    Phase 0: Kickoff and Data Confirmation (Next 48h)

    • Objectives: Align owners, confirm folder structure/secure access, and agree on confidentiality protocols.
    • Activities: 30–45 minute kickoff meeting; creation of secure folders (BlueHarbor/Technical, BlueHarbor/Procurement, BlueHarbor/Stakeholder) with RBAC.
    • Deliverable: Dataset receipt log and list of blockers.

    Phase 1: Rapid Diagnostic (Day 1–5)

    • Activities: Data validation/cleaning; preliminary load/diesel exposure, outage clustering, and siting constraints; stakeholder interviews.
    • Deliverable: 1-page diagnostic per island + consolidated data appendix.

    Phase 2: Modelling & Scenarios (Week 2–3)

    • Objectives: Produce scenario analysis for three modular options: Small (partial battery/inverter), Medium (hybrid with genset control), and Full (diesel displacement).
    • Activities: Demand scenario construction; technical simulation (hourly dispatch modelling); financial modelling (CAPEX/OPEX, financing scenarios); sensitivity analysis (fuel price, FX, lead time, etc.).
    • Deliverables: Scenario deck with quantified outcomes and investor-facing summary of financing conditions.

    Phase 3: Recommendations & Implementation Plan (Week 4)

    • Objectives: Convert modelling into a pragmatic implementation plan and procurement approach.
    • Activities: Site prioritization; procurement strategy (RFP templates, vendor shortlists); 30/60/90 day implementation timeline.
    • Deliverable: Recommendation memo with implementation sequencing and procurement annex.

    Phase 4: Investor-grade Business Case (Week 5–6)

    • Objectives: Finalize documentation for lenders/donors.
    • Activities: Refine CAPEX/OPEX and LCOE ranges; draft term-sheet/financing memo; prepare procurement/spec annex.
    • Deliverable: Investor-grade business case and term-sheet ready package.
  12. Review the GreenRiver Rapid-Response Deliverables and Timeline

    main

    The GreenRiver Rapid-Response plan follows a strict timeline starting from T+0 (kickoff). Key milestones include:

    • T+0–6 hours: Data intake and site feasibility assessment.
    • T+6–24 hours: Technical note delivery and GIS overlay initiation.
    • T+24–48 hours: Needs-assessment memo, site shortlist, and clinical SOP completion.
    • T+48–72 hours: Finalized needs-assessment and concept directions.

    Key Deliverables by Owner:

    • Lisa: Kickoff facilitation, data ingestion, 48h needs-assessment memo, site shortlist, and procurement priority list.
    • Alex: Data ingestion, GIS overlay & access map.
    • Oscar: 24h technical note (cold-chain/shelter), pilot site constraints, and procurement priority list.
    • Marcos: Clinical SOPs, AEFI checklist, and community engagement scripts.

    All times are referenced to local BRT/CET. The proposed kickoff is 07:30 BRT (11:30 CET) or 18:30 BRT (22:30 CET).