FINAGG INNOVATION LAB · BUILD 2026.05 v 4.2.0 · INDIA

Nine lending agents.
One reasoning layer.

FinAGG builds LendingAgents — the reasoning layer that sits between bureaus, account aggregators, GSTN, bank statements and your underwriting model. A suite of nine specialised agents that read documents, score risk, trace skips, monitor portfolios and answer in plain language.

finagg ▸ agent-trace ▸ session.j8w2
live p99 · 412ms
Files processed · 30d
8.42M
Median decision latency
412ms
Lenders in production
37+
LAUNCHING LendingAgents.AI  ·  9 agents to make MSME lending effective
MON 18 · 05 · 2026 14:00 IST lendingagents.ai
// IN PRODUCTION AT Largest NBFC, Largest Private Sector BANK, Largest SFB // +29 LENDERS
THE LAB
§ 01 / 05

We build agents that make decisions, not chatbots that recommend them.

The FinAGG Innovation Lab is a small group of researchers and credit veterans shipping financial agents into regulated production. Every product below started here as a single, sharp problem.

Read what humans read.

Agents must work with the actual artefacts of Indian credit — GSTR-1B, bank statements, ITR, RoC filings, KYC photos — not pristine API payloads.

Decide, then explain.

Every output is a structured decision with a citation back to the source line. A risk officer should be able to audit any field in seconds.

Latency is a feature.

Underwriting at scale means sub-second p99. We co-design models and retrieval so decisions arrive while the customer is still on the screen.

Composable, not monolithic.

FAME Score, CIC, OCR, AA, SKIP — each is a standalone agent with a clean API. Run one, run all, run them inside your own VPC.

Built as per compliance defined by regulator.

SOC 2 Type II, ISO 27001, RBI DPDP-aligned. Data residency in India. Model cards, eval suites and red-team logs available on request.

Boring on the outside.

The interesting thing about a financial agent is that nothing surprising happens. We measure ourselves in incident-free quarters.

PRODUCT SUITE
§ 02 / 05

Nine LendingAgents. One reasoning layer.

Each card opens a focused agent in the LendingAgents suite — hover to see the live shape of its output, the actual JSON, table, score or trace it returns to your stack.
lendingagents.ai ↗  lendingagents.in ↗

01 · SCORING FAME Score // GA
v 4.2

Cash-flow underwriting score for thin-file Indian borrowers. Trained on 11M AA+GST+bank profiles, calibrated against 36-month repayment outcomes.

subject · 28FXXPS9712E1ZWPASS
CASH-FLOW 742 / 900
DSCR 1.82 Vintage 24m GST ratio 0.94
famescore.in
02 · BUREAU Credit Intelligence // GA
CIC

Multi-bureau CIBIL / Experian / CRIF parsing with deduplication, alias-merge and tradeline-level alerting in one normalised schema.

CIBIL
786
EXPERIAN
771
CRIF
779
tradelines · 14 active · 2 disputed
enquiries · 7 / 90d
/cic
03 · COPILOT FAME GPT // GA
copilot

Ask the underwriting agent. Plain-language reasoning over a borrower's full credit + cash-flow + GST footprint, with citations to source lines.

RM · Anjali
"why is this score below cutoff?"
DSCR dropped to 0.74 in Apr–Jun [BS pg 4] after a ₹2.1L vendor outflow. 3 EMIs delayed in 6m [CIC tl-09].
/AskFAME
04 · DATA AA with GSTN // GA
consent

One consent flow, two channels. Pull AA bank statements and GSTR returns together, reconciled into a single working-capital view.

consent · ACT-7F2C-9B
✓ AA · 11 banks ✓ GSTR-1 · 24m ✓ GSTR-3B · 24m
reconciled in 3.2s
/gstnAa
05 · INGEST Document Analysis & OCR // GA
vision

Vision-LM tuned on 14M Indian credit documents — bank statements, ITR, GST certs, Form-16, PAN, Aadhaar — outputs structured fields, not text dumps.

documentBS · HDFC · pg 1–8 fields extracted142 / 142 tampering checkPASS · 0.001 closing bal₹ 4,82,109.22
/section
06 · POST-BOOK Portfolio Monitoring // GA
real-time

Continuous risk surveillance on the book — GST defaults, bounce events, new enquiries, cohort decay — pushed into your collections workflow as alerts.

alerts · last 24h
12 · GSTR-3B miss (T-3)
41 · bounce events
28 · new bureau enquiries
/portfolio
07 · ORIGINATION Agentic Onboarding // GA
e2e

A single agent runs the full origination: KYC, AA pull, GST pull, OCR, fraud, FAME Score, policy check — branched flows your ops team can edit.

KYC AA GST OCR SCORE POLICY
elapsed 2m 14s · 0 human touchpoints
/onboarding
08 · COLLECTIONS SKIP Tracing // BETA
beta

Find delinquent borrowers without breaking DPDP. Agents reconcile EPFO, RoC, mobile networks, e-commerce footprints into a current address graph.

subject · DPD 92
↳ EPFO · employer change [Aug '25]
↳ RoC · new directorship [Jan '26]
↳ delta-addr · Whitefield, BLR
/skip
09 · VERIFICATION AI Personal Discussion // BETA
beta

A voice agent that conducts the PD call in Hindi / English / Tamil / Marathi, asks the policy-driven follow-ups and writes the case note.

call · 04m 12s · hi-IN
A: "आपकी मासिक income ₹85k दिख रही है — confirm?"
B: "हाँ, करीब उतनी ही।"
→ income consistency PASS
/pd
LIVE DEMO · FAME GPT
§ 03 / 05

Ask the underwriting agent anything.

A real session against an anonymised SME applicant. Try a question — answers stream back grounded in the borrower's bureau, AA, GST and OCR'd documents.

FAME GPT grounded · 14,221 tokens of context session · 7c3a-9f
RM
RM Anjali · CR-3122
Vikrant Traders applied for ₹40L working-capital. Anything I should worry about?
FA
FAME GPT · v4.2
Three things stand out.

1. GST cadence is solid — GSTR-3B filed on time in 23 of 24 months [gstn · summary]. Output tax has grown 31% YoY.

2. Bureau is clean but recent — primary applicant pulled 3 enquiries in the last 60 days [cic · enq]; one was an unsecured BL of ₹12L from Q-Bank not yet visible in AA cashflows [bs · pg 14].

3. One DSCR dip in Apr–Jun '25 from a ₹2.1L vendor settlement [bs · pg 4]. Cash position recovered within 6 weeks.
calls · cic.tradelines() · aa.balance(11) · gstn.returns(24)
RM
RM Anjali · CR-3122
Recommend amount and tenor.
FA
FAME GPT · v4.2
FAME · RISK LOW · B+
AT SCALE
§ 04 / 05

Production-grade since day one.

We run inside regulated balance sheets, so the boring stuff matters first. SLAs, residency, audits — same as your core banking vendor.

Loans decisioned
₹14,200cr
cumulative across the suite, FY25–26
Documents read
94.7 M
98.4% straight-through extraction
SLA · p99
< 900ms
end-to-end agent response
Uptime · 12m
99.98%
incident-free quarters · 4 / 4
COMPLIANCE

SOC 2 · ISO 27001 · DPDP-ready

Audited annually by KPMG. All processing in Indian regions (Mumbai, Hyderabad). BYOK supported. Customer-managed VPC deployment available for lenders > ₹5,000cr AUM.

SOC 2 · Type II ISO 27001:2022 DPDPA aligned RBI · DLG
ARCHITECTURE

Three layers. One contract.

data → Account Aggregator · GSTN · Bureaus · KYC sources, ingested once.
reasoning → Specialised agents for OCR, scoring, tracing, copiloting.
surface → REST · webhooks · in-product copilot · or your own LOS.

WHAT'S NEXT
§ 05 / 05

Inside the Lab.

Research bets we are running now. Some will become products, some will quietly go into existing agents, some will be wrong. We publish post-mortems either way.

#ProjectStageOwnerETAStatus
R-014Multilingual PD agent (10 IN languages)EvalSpeech · NiharikaQ2 '26green
R-021Bharat KYC v2 — Aadhaar masked, offlinePilotVision · RahulQ3 '26green
R-029Real-time GSTR-1 anomaly streamingBuildData · SnehaQ2 '26yellow
R-031Settlement-grade SKIP graphResearchGraph · DevanshQ4 '26yellow
R-034Agentic dispute resolution (RBI OBO)DiscoveryPolicy · Aartired
Build the next decision
on FinAGG.

We work with banks, NBFCs and fintechs running > 25,000 originations a month. Production pilots ship in under 6 weeks.