Forensic intelligence that never leaves the room

ChanakAI reconstructs the money trail from bank statements, interrogates case documents with citations, and drafts the reports an officer actually files. Every inference runs on the local machine. There is no cloud call to make.

Running on Gemma 4 12B Qwen 3.5 9B your own GGUF
Works with no internet connection Signed, tamper-evident audit log Startup India, DIPP116733
Works with
PDFScanned PDF + OCRDOCXXLSXPPTXEMLCSVGemma 4 12BQwen 3.5 9BAny GGUF modelHindi + EnglishDevanagari + Gujarati PDFScanned PDF + OCRDOCXXLSXPPTXEMLCSVGemma 4 12BQwen 3.5 9BAny GGUF modelHindi + EnglishDevanagari + Gujarati PDFScanned PDF + OCRDOCXXLSXPPTXEMLCSVGemma 4 12BQwen 3.5 9BAny GGUF modelHindi + EnglishDevanagari + Gujarati

The money trail

Bank statements become a transaction ledger, then a stage-by-stage flow of funds with accounts ranked by how recoverable they still are.

The case file

PDFs, scans, spreadsheets and email answered in plain language, with every figure pointing back to the page it came from.

The report you file

Findings drafted into the formats an officer submits, exported to PDF or DOCX with an optional chain-of-custody appendix.

0Outbound calls
16 GBMinimum RAM
AnyOpen model you bring
100%Processed on device

A financial case, end to end.

Statements in, money trail out, and the report filed. Every step runs on the officer's own machine.

From statements to a money trail

Any bank, any layout. The columns are mapped, the arithmetic is checked, and the flow of funds comes out the other side.

Reading evidence
Open Case 337 / Statements
PDFhdfc_4471_apr-sep.pdf2.4 MB
XLSXsbi_2038_statement.xlsx840 KB
PDFicici_9155_scanned.pdf6.1 MB
PDFaxis_7719_apr-sep.pdf3.2 MB
CSVncrp_complaints.csv312 rows
XLSXvictim_statements.xlsx96 KB
TXTnotes.txt4 KB
3 files selected Open
PDFhdfc_4471_apr-sep.pdf412 rows
XLSXsbi_2038_statement.xlsx305 rows
PDFicici_9155_scanned.pdfOCR · 567 rows
1,284 transactions read · nothing left the machine
Holder R. K. Deshmukh A/C ****4471 Period Apr–Sep 2026 Rows 412
Column in fileMapped toFit
TarikhTransaction date99%
VivaranNarration97%
NaameDebit94%
JamaCredit94%
BakiRunning balance96%
Chq / RefInstrument reference92%
Mulya TithiValue date91%
ShakhaBranch88%
Balance continuity holds · totals tie · auto-applied
Which accounts in case 337 are still freezable?
Understood: which accounts still hold balance
Planned 3 sub-questionsdeep
Reasoned across 1,284 transactions
Reasoned over 1,284 transactions

Three accounts still hold recoverable balance, ₹41L of the ₹2.4cr traced:

HDFC ****4471: ₹18.4L of ₹22L received, no debit since 11 June1

SBI ****2038: ₹6.2L of ₹15L, partially withdrawn2

Axis ****7719: ₹16.4L of ₹19L, no outward transfer3

Citations [1] hdfc_4471_apr-sep.pdf · p14 [2] sbi_2038_statement.xlsx · row 812 [3] axis_7719_apr-sep.pdf · p6
Type a message… Which accounts in case 337 are still freezable?
Think qwen3.5 9B

Ask it to show you

The same mapped ledger, answered three ways: as a flow, as totals by layer, and as where the money actually ended up.

Drafting report
Show me the money flow for this case
Classified every account to a chain stage
Aggregated 1,284 transactions stage to stagedeep
Nothing untagged left out of the totals
Traced 5 layers across 37 accounts

₹2.4cr moved through 5 layers in 11 days:

Citations [1] 1,284 mapped transactions
Type a message… Show me the money flow for this case
Think qwen3.5 9B
How much moved at each layer?
Summed by chain stage
Excluded duplicates and reversalsexact
Summed 1,284 transactions

Layer 2 carries more than layer 1, which is the re-pooling:

Victims₹2.4cr
Layer 1₹1.6cr
Layer 2₹2.1cr
Cash-out₹1.9cr
Citations [1] forensic_txn · 1,284 rows
Type a message… How much moved at each layer?
Think qwen3.5 9B
Where did the money end up?
Split the exit layer by disposal
Checked closing balances against statementsexact
Split the exit layer three ways

Most of it left as cash, which is why the ₹41L matters:

Withdrawn as cash₹1.9cr
Still held, freezable₹41L
Charges and reversals₹9L
Citations [1] closing balances · 3 statements
Type a message… Where did the money end up?
Think qwen3.5 9B

Or just a folder of documents.

No statements, no ledger. Drop in what the case has accumulated, ask it questions, and file what you find.

Ask the case file anything

Questions in English or Hindi, answered from the material you ingested, with every claim carrying the document and page behind it.

Reading evidence
Drop files to add them to this project
PDFfir_0142.pdf14 pages
DOCXfsl_examination.docx8 pages
EMLnodal_officer_thread.eml31 messages
PDFfir_0142.pdf86 chunks
DOCXfsl_examination.docx42 chunks
EMLnodal_officer_thread.eml57 chunks
185 passages indexed · entities extracted on device
What is still outstanding on this case?
Understood: what has no recorded response
Planned 2 sub-questionsdeep
Read 185 passages for open threads
Reasoned over 185 passages

Three items in this case have no recorded response:

FSL report: relied on in the FIR, never placed on file1

Nodal officer request: sent 12 May, unanswered in 31 messages2

Call records: requested 19 May, not received3

Citations [1] fir_0142.pdf · p2 [2] nodal_officer_thread.eml · msg 18 [3] nodal_officer_thread.eml · msg 26
Type a message… What is still outstanding on this case?
Think qwen3.5 9B

Drafted, then sent

The same answer written up into the report you submit, and delivered without leaving the workstation.

Drafting report
Template Investigation findings Version v2 Pages 6
Case summary
Chronology of events
Documents relied upon
Outstanding actions
DOCXPDFHTMLMD
Redact identifiersOn
Toio.station12@gov.in
SubjectFIR 0142 · outstanding actions
AttachedDOCXfir0142_review.docx
Delivered · logged to the custody chain

And when the files only make sense together.

Understand reads a set of related documents as one connected map, so a name that appears in four of them becomes a single point that ties them.

Woven into one network

Each file is read for its key ideas, then the concepts that appear in more than one become the bridges that hold the case together.

Reading evidence
complaint_bundle.pdfPDF · 22 p
device_seizure_log.xlsxXLSX · 41 rows
bank_correspondence.emlEML · 18 msg
station_diary.pdfPDF · 9 p
call_records.csvCSV · 2.1k
notes.txtTXT · 4 KB
3 of 6 selected Add to workspace
PDFcomplaint_bundle.pdfready
XLSXdevice_seizure_log.xlsxready
EMLbank_correspondence.emlready
Key ideas pulled from each file · ready to weave

Then explained, one concept at a time

Pick any point on the map and it tells you what it is, which files it came from, and what it connects to.

Explaining a concept
Handset IMEI 3556…091In 3 files
Named in the complaint as the caller's handset
Recovered in the seizure log, item 4
Used to register the beneficiary account
Seen again in the bank correspondence thread
Appears in 3 of the 3 files you added
First seen in the complaint, 2 May
Beneficiary accountIn 2 files
Opened with the same handset IMEI
Named in the bank correspondence thread
Receives from both primary mule accounts
No onward transfer recorded after 14 May

One identifier ties the complaint, the seizure and the bank trail into a single account of what happened.

From a folder of evidence
to a filed report.

01

Bring the case material in.

Statements, FIRs, lab reports, spreadsheets and email. Scanned pages go through OCR. Nothing is uploaded anywhere, because there is nowhere to upload it to.

02

Trace the money and rank the freezes.

Statements are mapped, validated against balance continuity and declared totals, then traced through the chain. Accounts are ranked by what is still recoverable, because recoverability decays by the hour.

03

Draft it, check it, file it.

The model drafts, the officer attests. Reports are versioned, exportable and can carry their own handling history as an appendix. AI never produces a signature or an attestation.

Custody is a product
surface, not backend hygiene.

Local inference is the premise. What it buys you is that the handling history of a case is complete, local and tamper-evident, and can be printed as part of the document that carries the finding.

Hash-chained and signed

Each entry is signed and linked to the previous one, with the key sealed to the Windows user on that machine.

Printable as an appendix

Export a report with its own handling history attached, each row showing the chain hash that ties it to the signed log.

Verifiable independently

A verifier re-walks the whole file and checks the chain and every signature, without the application running.

Nothing to leak

No telemetry leaves the device and no third-party asset is fetched at runtime. Everything ships with the install.

Bring your own model.

Nothing here is tied to one vendor's weights. Add any GGUF model from a file or a URL, and the registry tells you whether your card can actually run it before you commit.

Add by file or by URL

Pick a GGUF off the disk, or paste a URL and watch the download progress. Shipped entries stay read-only; fork one to make it editable.

Told before you commit

VRAM is estimated against the card actually in the machine, a sibling vision projector is detected automatically, and a smoke test loads it for real.

Swapped without losing work

Per-model offload, context size and flash-attention settings. Reloading the router takes seconds and open conversations survive it.

Ships configured with Gemma 4 12B and Qwen 3.5 9B (with a vision projector), plus separate embedding and reranking servers. CPU-only machines work; a GPU makes it faster.

Built for the work
that actually lands on the desk.

Money-trail reconstruction

Multi-bank statements into one ledger, layered flows surfaced, accounts ranked by freeze priority.

Cyber / EOW

Case file interrogation

A whole case file read at once: persons, evidence inventory, chronology and the gaps still outstanding.

Investigation

Statutory drafting

Case diary entries, seizure memos, electronic-evidence certificates and chargesheet annexures, drafted and versioned.

Pending vetting

Connecting related files

Drop a set of related documents in and see the concepts that bridge them, as one connected map.

Intelligence

Hindi and English working

Translation and audio transcription on device, with Devanagari and Gujarati rendering throughout.

Government

Registers and spreadsheets

Ask a station diary, a procurement register or an audit workbook a question and get an answer tied to rows.

Government

The things people ask first.

Does any case data leave the machine?

No. Models, index and case data all sit on the workstation and inference runs locally, so there is no cloud call to make. The one optional network feature is sending a report by email, which uses your own department SMTP server on your own network, and is queued locally if that server is unreachable.

What hardware does it actually need?

A standard Windows workstation with 16 GB of RAM is the floor, and it runs on CPU alone. A GPU makes generation faster; the hardware panel reports your card, its VRAM and whether it meets the compute capability the CUDA build needs.

Can the output be used in court?

Not as it stands. The statutory templates have not yet been through legal vetting and ship behind a notice saying so. Everything the model produces is a draft an officer must verify and attest to. ChanakAI never generates a signature or an attestation, by design.

How do I know the AI has not mis-read a statement?

Because a mapping is not accepted on the model's confidence. It has to survive checks that are arithmetic rather than linguistic: the columns must exist, the dates and amounts must parse, running balances must reconcile, and where the bank prints its own totals, ours must equal them. If a check fails, the officer is shown which column is wrong and why.

Which models can we run?

Any GGUF model you choose to bring, added from a file or a URL. It ships configured with Gemma 4 12B and Qwen 3.5 9B, but neither is a lock-in: fork the shipped entry, point it at other weights, and reload.

See it run on the laptop
you already have.

A live demonstration on a standard 16 GB Windows machine. No special hardware, no cloud account, no data of yours involved.

Procurement, tender or GeM enquiries: hello@karpuragaurai.com