ReguLattice

• 

  Salary Slabs

• 

  Report

• 

  Settings

• 

  Evidence Vault

• 

  Dashboard

• 

  Task Management

• 

  Mapping Evidence with Policy

Inspiration

ReguLattice is inspired by the massive friction in tax compliance for emerging markets — especially Pakistan’s FBR system, where manual processes, missed savings (zakat/donations/pension), withholding errors, and audit gaps cost SMEs and salaried individuals billions annually. Gemini 3’s 1M token context, multimodal reasoning, and agentic capabilities allow us to move from static tools to autonomous, evidence-grounded compliance agents that plan, validate, and self-correct over long workflows — turning regulatory nightmares into intelligent, proactive systems.

What it does

ReguLattice automates and optimizes tax compliance for businesses and individuals under FBR regulations (expanding to UAE, KSA, USA). It uses Gemini 3 Pro + RAG to:

Calculate salary tax in real-time with slab analysis and savings credits Verify withholding via a 10-step guided process (filer/non-filer logic) Validate uploaded evidence semantically against regulations Generate tasks, score compliance, detect gaps, forecast penalties Build audit-ready evidence packs with traceable reasoning

The agent plans multi-step workflows (retrieve → classify → reconcile → draft → self-check), maintains continuity across sessions, and provides grounded outputs with full evidence trails — all while keeping humans in the loop for final commitments.

How we built it

Core stack: React/TSX frontend, Supabase/pgvector backend, Gemini 3 Pro for reasoning and multimodal understanding RAG pipeline: Gemini 3 ingests long-context regulatory docs, payroll CSVs, challans, and bank statements; retrieves relevant rules and evidence, reasons over them to validate matches and detect discrepancies Agentic layer: Gemini 3 plans and executes multi-step tasks (e.g., evidence retrieval → classification → narrative draft → consistency check) using Thought Signatures and Thinking Levels for continuity and self-correction Guardrails: Bounded agency — restricted data sources, no auto-filing, mandatory evidence + human sign-off Built solo: Full MVP coded by me (Moazzam) — from UI to RAG orchestration

Challenges we ran into

Long-context reasoning limits on earlier Gemini versions — solved by Gemini 3 Pro’s 1M token window, enabling full payroll + regulatory doc analysis in one pass Probabilistic variance in RAG outputs — addressed with structured intake, golden dataset testing, and mandatory evidence traceability Balancing agent autonomy with regulated safety — implemented strict guardrails (no write/submit, always human approval) to prevent “automation theatre” Multi-jurisdiction complexity (Pakistan FBR → UAE FTA → KSA ZATCA) — built dynamic RAG index with jurisdiction boundaries to prevent cross-border leakage

Accomplishments that we're proud of

Built complete MVP solo in 2 months: full React frontend, Supabase/pgvector backend, Gemini RAG pipeline for evidence validation and compliance scoring Incubated at National Incubation Center Hyderabad (2025 cohort-7) Published research on quantum-enhanced RegTech and ethical AI governance (Global Business & Economics Journal, Hampton Global Business Review) Media features in Success Quarterly and TechTimes for responsible AI work

What we learned

Gemini 3’s multimodal reasoning + long context dramatically reduces manual evidence gathering — but guardrails and human-in-the-loop are non-negotiable in regulated domains Probabilistic agents excel at evidence-heavy, long-running tasks (audit packs, gap detection) but must never own final calculations or submissions Emerging-market tax compliance is high-friction and high-stakes — AI can unlock massive value when grounded in local rules and real evidence Solo building forces ruthless prioritization and deep ownership — every feature is battle-tested by necessity

What's next for ReguLattice

Agentic evolution: Use Gemini 3 Pro’s Thought Signatures and Thinking Levels to build full Marathon Agents that autonomously orchestrate monthly compliance cycles (retrieve evidence → validate → score → draft report → self-correct → flag for human review) Multi-jurisdiction expansion: Add UAE FTA VAT, KSA ZATCA e-invoicing, and USA payroll/sales tax rules to the RAG index Enterprise integrations: Connect to ERP APIs (Odoo, Sage) and bank statements for real-time data ingestion Audit autopilot: Fully autonomous audit pack generation with Gemini 3 multimodal understanding (scan receipts, documents) Global scaling: Target diaspora SMEs in USA, high-growth SMEs in GCC, and multi-country businesses in Southeast Asia

Built With

• deno
• gemini
• google-drive
• html5
• javascript
• react
• sql
• supabase
• tailwind
• typescript