Milaan
Milaan
Autonomous Bipartite Solver · Buildathon
000%
Caching Graph Nodes & Synthetic Batch Feeds
INIT PROTOCOL: FELLEGI-SUNTER & KUHN-MUNKRESSYNTHETIC 60–80 RECORD BATCH
Editorial financial texture primary
Editorial financial texture secondary
Editorial financial texture primary
Editorial financial texture secondary
Editorial financial texture primary
Editorial financial texture secondary
Editorial financial texture primary
Editorial financial texture secondary
Editorial financial texture primary
Editorial financial texture secondary
AI-Powered Reconciliation · Razorpay Buildathon Track 04

Reconciling

Probabilistic Fellegi-Sunter scoring + LLM adjudication over noisy synthetic bank transaction narrations

the Rupee

Modeling asymmetric TDS §194C/J deductions, GST net invoices, and UTR clearing

Ledger to Ledger

Kuhn-Munkres O(n³) Hungarian assignment resolving greedy allocation collisions

AI-Adjudicated

Gemini 2.5 Flash-Lite → Groq Llama 3.3 70B failover explains every ambiguous match in plain English with a bounded ±20 confidence delta

Explained

Verifiable confidence scores across 60–80 synthetic records — AI enhances explanation, deterministic code enforces the final decision

Accountant bahi-khata ledger desk
Digital transaction routing flow
Balanced reconciliation archival ledger
02 / THE RECONCILIATION PROBLEM

Reconciliationbreaksthemomentrealmoneymoves.

TruncatedRTGSnarrations.AsymmetricTDSdeductionsunder§194C&§194J.Missinginvoicenumbersacrosssplitbatchsettlements.

MilaaneliminatesgreedycollisionswithprovableHungarianoptimization.

SCROLL PROGRESSIONPROVABLE RESOLUTION
PASS 01Deterministic Rule Matching
PASS 02Fellegi-Sunter Probabilistic Linkage
PASS 03Semantic & Phonetic Vendor Matching
PASS 04Combinatorial Optimization
PASS 05LLM Adjudication
PASS 01Deterministic Rule Matching
PASS 02Fellegi-Sunter Probabilistic Linkage
PASS 03Semantic & Phonetic Vendor Matching
PASS 04Combinatorial Optimization
PASS 05LLM Adjudication
PASS 01Deterministic Rule Matching
PASS 02Fellegi-Sunter Probabilistic Linkage
PASS 03Semantic & Phonetic Vendor Matching
PASS 04Combinatorial Optimization
PASS 05LLM Adjudication
SECTION 194C TDS
SECTION 194J TDS
GST RULE 36(4) ITC
UPI AUTOPAY CLEARING
NEFT & RTGS SETTLEMENT
IMPS INSTANT LEDGER
AS-9 REVENUE ACCRUAL
SYNTHETIC BENCHMARK HARNESS
AI CONFIDENCE SCORING
GEMINI 2.5 FLASH-LITE
LLM EXPLAINABILITY
GROQ LLAMA 3.3 70B
SECTION 194C TDS
SECTION 194J TDS
GST RULE 36(4) ITC
UPI AUTOPAY CLEARING
NEFT & RTGS SETTLEMENT
IMPS INSTANT LEDGER
AS-9 REVENUE ACCRUAL
SYNTHETIC BENCHMARK HARNESS
AI CONFIDENCE SCORING
GEMINI 2.5 FLASH-LITE
LLM EXPLAINABILITY
GROQ LLAMA 3.3 70B
SECTION 194C TDS
SECTION 194J TDS
GST RULE 36(4) ITC
UPI AUTOPAY CLEARING
NEFT & RTGS SETTLEMENT
IMPS INSTANT LEDGER
AS-9 REVENUE ACCRUAL
SYNTHETIC BENCHMARK HARNESS
AI CONFIDENCE SCORING
GEMINI 2.5 FLASH-LITE
LLM EXPLAINABILITY
GROQ LLAMA 3.3 70B
03 / ALGORITHMIC EVALUATION BENCHMARKS

Synthetic Batch Performance

Evaluated on synthetic test batches modeled after Indian banking narrations and statutory TDS deduction structures.

01Evaluation Batch Size
60–80

Synthetic Test Records

Evaluated on simulated Indian bank feeds and ERP rows containing intentional noise, truncated RTGS narrations, and split payments.

02O(n³) Bipartite Latency
0.38ms

Hungarian Solve Time

Calculates minimal-cost bipartite optimal assignment matrix to eliminate greedy allocation collisions across concurrent rows.

03Production Benchmark
—

Pending Evaluation Run

Formal benchmark against live production enterprise datasets will be conducted post-buildathon evaluation.

04 / THREE-PILLAR CONVERGENCE

The Core Triad

Three algorithmic pillars converge into one deterministic confidence score.

Rules Engine
PILLAR 01

Rules Engine

Deterministic Invariants

PAN-UTR exact pairing, statutory TDS §194C/J threshold brackets, and bank IFSC routing tables.

Fuzzy + Semantic
PILLAR 02

Fuzzy + Semantic

Probabilistic Matching

Fellegi-Sunter log-likelihood vectors, sentence-transformer embeddings (all-MiniLM-L6-v2), phonetic fuzzy resolution — ambiguous cases escalated to LLM adjudication.

Combinatorics
PILLAR 03

Combinatorics

Bipartite Optimization

Kuhn-Munkres O(n³) Hungarian assignment and subset-sum solvers resolving multi-invoice bulk payouts.

100% RECONCILED
LLM-Explained · Bounded Confidence
KUHN-MUNKRES MATCHING ENGINEZERO COLLISION PROBABILITY
06 / AI ADJUDICATION LAYER

The Machine That Explains Itself

Pass 5 invokes an LLM only on genuinely ambiguous records. AI enhances explanation quality. Deterministic code enforces the final decision.

GOOGLE DEEPMIND
GEMINI 2.5 FLASH-LITE

Primary AI Adjudicator

Invoked only on medium-confidence ambiguous records. Proposes a plain-English explanation + a bounded confidence delta. Never makes the final accept/reject call alone.

GROQ INFERENCE
GROQ LLAMA 3.3 70B

Automatic Failover

Seamless failover when Gemini rate-limits mid-batch. Exponential backoff, circuit breaker, and graceful degradation ensure the batch completes even if both LLM providers are exhausted.

AUDIT TRAIL
SCHEMA-VALIDATED AUDITED

Bounded and Explainable

Every LLM response is validated with Pydantic, stored as raw_llm_response in the audit log, and capped at a bounded confidence delta. Hallucinations cannot silently corrupt the ledger.

Design principle: AI proposes. Deterministic code decides. LLM confidence adjustments are strictly bounded. The final accept / review / reject threshold is enforced by auditable, inspectable code. Hallucinations cannot silently pass a bad match.

05 / MATHEMATICAL BENCHMARKS

Evaluation Metrics

SYNTHETIC TEST BENCH
METRIC 01
0 Records
Evaluation Batch
Simulated Indian RTGS, NEFT, and IMPS synthetic records
METRIC 02
0.00ms
Solve Latency
Per Kuhn-Munkres O(n³) Hungarian bipartite matrix step
METRIC 03
0 Passes
Algorithmic Pipeline
Rules · Fuzzy · Embeddings · Combinatorics · LLM Adjudication
METRIC 04
—
Production Benchmark
Formal live enterprise dataset evaluation pending post-buildathon
Manual Spreadsheet Guesswork→Autonomous Bipartite SolverO(n³) Kuhn-Munkres Polish
Manual Spreadsheet Guesswork→Autonomous Bipartite SolverO(n³) Kuhn-Munkres Polish
Manual Spreadsheet Guesswork→Autonomous Bipartite SolverO(n³) Kuhn-Munkres Polish
Manual Spreadsheet Guesswork→Autonomous Bipartite SolverO(n³) Kuhn-Munkres Polish
Manual Spreadsheet Guesswork→Autonomous Bipartite SolverO(n³) Kuhn-Munkres Polish
Manual Spreadsheet Guesswork→Autonomous Bipartite SolverO(n³) Kuhn-Munkres Polish
Milaan

An autonomous financial reconciliation prototype exploring probabilistic Fellegi-Sunter scoring, Kuhn-Munkres Hungarian bipartite matching, and Benford's Law anomaly detection across synthetic Indian financial datasets.

SECTION 194C · SECTION 194J · GST RULE 36(4) · SYNTHETIC EVALUATION
Run Evaluation Batch

Dispatch a synthetic test batch in seconds.

Deploy Intake Dispatch
Tax Logic Modeled
  • TDS §194C (1% / 2%)
  • TDS §194J (10%)
  • GST Rule 36(4) Matching
  • Net vs Gross Analysis
Stack & Algorithms
  • Kuhn-Munkres O(n³)
  • Fellegi-Sunter EM
  • Benford Chi-Square
  • Next.js + FastAP
Buildathon Project
© 2026 MILAAN RECONCILIATION · BUILDATHON PROTOTYPE