NOCTRODEX pitch deck — text version
Slide 1 — Title (Hero)
⚡ EDHTA · paper v3
// P U B L I C _ P I TC H · P R E - S E E D · AU G U S T 2 0 2 6
NOCTRODEX
( O P C ) P R I VAT E L I M I T E D
Two equal product stages under one company.
AI AppSec software now · sovereign on-prem hardware next
S TA G E 1 // S O F T WA R E
S TA G E 2 // H A R D WA R E
AI SAST & DAST
On-prem AI hardware
AppSec software assist
Code & apps stay under customer control
Appliances you own or manage
No per-query rent for inference
Architecture: EDHTA (paper v3) — in Stage 1 apps · designed for Stage 2 appliances
Kolkata, India · noctrodex-hq.com · Pre-revenue · Public · internals redacted
Slide 2 — One Thesis, Two Stages
One thesis, two stages
Software wedge first · sovereign compute next — full objects always include both
01
02
03
Prove value in AppSec
Own the inference path
Scale the stack
AI SAST + DAST assist where
code and apps stay under
customer control.
Same cyber AI delivered on
customer premises — no token
rent.
EDHTA architecture IP +
software + appliance SKUs +
managed path as capital allows.
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 2 / 25 ]
Slide 3 — Go-to-market Sequence
Go-to-market is staged — both stages are real
Commercial sequence, not abandonment of hardware (SBI / business overview still hold)
STAGE 1 — NOW
STAGE 2 — NEXT
AI AppSec software
On-prem AI hardware
AI SAST — static analysis assist on owned
code
AI DAST — authorised dynamic testing assist
Ghostmaster AI core (in development)
Operator / analyst software surface
Faster pilots · software IP · lower capital
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
Purpose-configured AI appliances
Local inference · no per-query metering
Outright sale or managed (ISP-router)
Models co-designed to hardware envelope
Sovereign compute for cyber + industrial
[ 3 / 25 ]
Slide 4 — Stage 1 Problem — Trilemma
Stage 1 problem — AppSec needs better AI
Cloud copilots leak code · one model window cannot do everything well
The pain
Proprietary code into public metered AI
Scanner noise vs late, expensive vulns
SAST misses runtime · DAST misses root
One model config cannot max long context +
deep judgment + tools
AppSec wants AI on SAST + DAST — with ownership
of source and scope.
Why one model window isn’t enough — EDHTA splits reader vs judge.
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 4 / 25 ]
Slide 5 — Stage 1 Products — AI SAST + DAST
Stage 1 products — AI SAST + AI DAST
Active development / pilot path · not public GA
AI SAST
AI DAST
Static Application Security Testing — AI-augmented
Dynamic Application Security Testing — AI-augmented
Ingest customer-owned source / builds
AI-assisted finding prioritisation
Analyst-friendly review workflows
Built for AppSec + engineering SDLC
Keep code off public metered APIs
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
Authorised testing of running applications
AI-guided triage of dynamic findings
Pairs with SAST for root-cause linkage
Customer policy & scope always apply
No unauthorised scanning — ever
[ 5 / 25 ]
Slide 6 — Stage 1 Buyers & Focus
Stage 1 buyers
AppSec-first positioning (not revenue share)
Outcomes they want
10%
15%
35%
20%
Faster SAST triage
Authorised DAST, less noise
Source stays controlled
Later path to on-prem AI
20%
AppSec / engineering
Security consultancies
Research / labs
Industrial path→S2
Product cos (SDLC)
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 6 / 25 ]
Slide 7 — Stage 1 Workflows
Stage 1 workflows
Abstract paths — no internal tooling inventory
SAST path
01
02
03
04
05
Ingest code
Analyse
Prioritise
Review
Track fix
DAST path (authorised scope only)
01
02
03
04
05
Define scope
Exercise app
Collect signals
Triage AI
Remediate
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 7 / 25 ]
Slide 8 — Ghostmaster AI Core
⚡ EDHTA · paper v3
Ghostmaster — AI core (public view)
Cybersecurity-oriented model line · under development · not GA
What it is
AI for research and defensive workflows on infrastructure the
customer controls.
Powers analyst assist for AppSec — SAST and DAST triage —
without putting sensitive code on a public token API.
Powered by EDHTA: long-context reading + high-precision judgment
on your path.
Product layers
1. Ghostmaster (model)
2. Apps & software
· Operator console
· SAST / DAST focus
3. EDHTA architecture
(paper v3 · company IP)
4. On-prem hardware
(Stage 2)
Status: active iteration · not public GA.
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 8 / 25 ]
Slide 9 — EDHTA — The Architecture We Ship
⚡ EDHTA · paper v3
EDHTA — the architecture we ship
Formal ML paper (v3) · company IP · Stage 1 active · Stage 2 designed to host
RAW
INPUT
READER
TYPED
BRIEF
JUDGE
ADMITTED
Code · apps
reports
Long context
to structure
Claims +
confidence
High-precision
AppSec assist
Findings
when gated
How AppSec work becomes a brief — not a raw dump into a single window.
Stage 1 — active
Stage 2 — designed
Company IP
In softwares & apps: Ghostmaster
path · AppSec assist · reader /
judge / memory.
Same architecture on owned onprem appliances — local inference.
EDHTA-v3 paper · copyright
package · not “patented” · not public
GA.
Full paper under company IP / NDA package · Document ID EDHTA-v3 · public view only
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 9 / 25 ]
Slide 10 — Stage 1 Architecture (EDHTA Map)
⚡ EDHTA · paper v3
Stage 1 architecture — EDHTA (public)
Shape map only · no ports · same design Stage 2 hosts on-prem
GATEWAY
READER
JUDGE
MEMORY
Routes long
vs short work
Long context
→ structured brief
High-precision
AppSec assist
Admitted findings
only
Same architecture Stage 1 software runs · Stage 2 appliances are designed to host it on customer-owned mind.
Long materials
Sharp judgment
Admitted memory
Code, reports, large pastes → briefs
Prioritise & assist without megawindow tax
Findings enter history only when
gated
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 10 / 25 ]
Slide 11 — Stage 1 Competitive Frame
Stage 1 competitive frame
Illustrative design intent — not a lab benchmark
AI depth
5
Position
On-prem path
Code privacy
Not “replace every scanner tomorrow.”
0
AI-native AppSec software that respects
ownership of code — with a clear Stage 2
hardware upgrade path.
Scores = design intent.
Noise control
Legacy SAST/DAST
S1+D link
Cloud AI copilots
Noctrodex S1 target
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 11 / 25 ]
Slide 12 — Stage 2 — Own The Compute
STAGE 2 // SOVEREIGN COMPUTE
Own the compute
On-prem AI appliances · co-designed models · no per-query metering
Where EDHTA runs on metal you own — equal brief: problem, product, deployment, buyers,
economics
Slide 13 — Stage 2 Problem — Renting Intelligence
Stage 2 problem — renting intelligence fails production
From business overview: economics · sovereignty · fit
SaaS never transfers control
5
DIY GPU ops burden
The missing category
4
Weak offline / air-gap
5
Data leaves the perimeter
5
Token costs scale with success
5
0
1
2
3
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
4
5
Turnkey AI appliances where the model
and the machine are designed together,
installed on the customer’s premises, and
fully operated by the customer.
Not raw GPUs. Not metered cloud.
6
[ 13 / 25 ]
Slide 14 — Stage 2 Solution — Co-Design
⚡ EDHTA · paper v3
Stage 2 solution — co-design model and machine
Local inference · EDHTA on owned hardware · not rent per query
Hardware
Purpose-configured AI appliances — compute, storage, power,
thermals.
Model stack
Domain-tuned models (incl. cyber) sized for the appliance envelope.
One architecture
Stage 1 software
EDHTA in apps
EDHTA architecture
Same paper as Stage 1 apps — reader / judge / memory on the mind.
Ops path
Stage 2 appliance
EDHTA on owned mind
Design → assemble → install → customer operates; air-gap capable.
Same paper. Your metal.
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 14 / 25 ]
Slide 15 — Appliance Classes
Stage 2 product surface — appliance classes
Thematic roadmap variants · not SKU quotes
Edge
Mid-range
High-capacity
Compact sovereign inference
for constrained sites and
offline cells.
Team / lab class stacks for
AppSec + research workloads.
Facility-class appliances for
heavier local model + tool
loops.
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 15 / 25 ]
Slide 16 — Stage 2 Deployment Models
Stage 2 deployment — outright vs managed
ISP-router analogy from business overview / SBI profile
No query metering
What never moves
On-prem compute
Compute stays on the customer’s
premises. Data does not leave for
inference. Nothing is billed per query.
Customer maintains*
What changes: title, capex vs opex, who
maintains the metal.
Capex preference
*Managed = Noctrodex maintains.
Outright = customer or contract.
Customer title
Illustrative scores.
0
1
Managed deployment
2
3
4
5
6
Outright purchase
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 16 / 25 ]
Slide 17 — Stage 2 Engagement Path
Stage 2 engagement path
How an appliance programme runs
01
02
03
04
05
Discovery
Co-design
Build
Deploy
Support
Workload, latency,
privacy, environment
Compute class,
memory, model
architecture
Assemble, validate,
load model bundle
Install on premises,
hand over control
Integration &
maintenance — not
access rent
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 17 / 25 ]
Slide 18 — Stage 2 Target Buyers
Stage 2 buyers
India first · industrial & institutional sovereignty (business overview §6)
Why they buy appliances
15%
25%
20%
20%
Data residency / offline
Predictable cost envelope
Own the metal and models
Graduate from Stage 1 software
20%
Manufacturing / OT
Healthcare on-site
SME / mid-market
Air-gap / critical
Research & education
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 18 / 25 ]
Slide 19 — Stage 2 Unit Economics
Stage 2 economics — planning view
BOM buckets illustrative · no supplier SKUs · re-quote before purchase
Unit economics (public)
15%
15%
45%
Margin from BOM + assembly + integrated
model value — not infinite cloud COGS per
query.
Managed deployments only when fee/advance
covers hardware outlay (bank profile honesty).
25%
Early company: capital discipline first.
GPU compute
CPU / memory / board
Storage & network
Power / chassis / other
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 19 / 25 ]
Slide 20 — Stage 2 Competitive Frame
Stage 2 competitive frame
Business overview alternatives — illustrative axes
Local control
5
Differentiation
Cyber fit
Cost predict.
Not a cloud API reseller.
Not a white-box GPU dump.
Not a pure body-shop SI.
0
Co-designed model + machine for
sovereign cyber and industrial AI.
Scores = design intent.
Turnkey
Cloud AI APIs
Air-gap
DIY GPU servers
Noctrodex S2 target
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 20 / 25 ]
Slide 21 — Company Revenue Mix
Company revenue mix — staged (illustrative %)
Pre-revenue · not a forecast of contracted business
120
Reading the chart
100
80
60
Y1* software-led (Stage 1).
40
Y2–Y3* hardware sale + managed
rise (Stage 2).
Axis = illustrative mix % of revenue
composition, not rupees.
20
0
Y1*
Y2*
SAST/DAST software
Hardware sale
Managed appliances
Support / training
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
Y3*
[ 21 / 25 ]
Slide 22 — Company Snapshot
Company snapshot
Public legal facts · street address omitted
6
5
5
4
4
3
Legal facts
5
4
NOCTRODEX (OPC) Pvt Ltd
Incorp: 27 Jul 2026
RoC: Kolkata, India
CIN: U62010WB2026OPC288993
NIC: 62010 · 62020 · 26201
3
2
Pre-seed · founder-funded
HQ: Kolkata, India
noctrodex-hq.com
1
0
SPICe
Incorp
Banking
S1 apps
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
S2 design
[ 22 / 25 ]
Slide 23 — Founder Background
Founder
Public site bio · education + professional background
Soumabha Majumdar
Founder & CEO · AI & Cyber Security Engineer
Dual prior career tracks: corporate software developer and infosec / security engineer.
Cyber forensics-trained AI security engineer. Professional experience spans corporate
product/engineering development and information-security engineering — including
security research, healthtech infosec, offensive research, and full-stack delivery —
before founding NOCTRODEX (OPC) PRIVATE LIMITED in 2026.
Corporate path
Infosec path
Education · certs
Prior corporate developer /
engineering experience
building software products
and full-stack systems.
Infosec engineer track:
security research, healthtech
infosec, authorised testing
craft.
MSc Digital Forensics (NFSU)
BTech · CEH · eWPTX
Kolkata, India
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 23 / 25 ]
Slide 24 — Stage Honesty & Roadmap
Stage honesty + roadmap themes
Build velocity · S1 then S2 · priority scores are illustrative (1–5)
₹0
0
Pre
S1→S2
External equity
Named clients
Revenue
GTM path
Near-term priority (1–5 illustrative)
AI SAST path
5
AI DAST path
5
AppSec pilots
5
Operator console
4
Appliance SKU (S2)
3
Managed deploy (S2)
3
// SYSTEM_STATUS: ONLINE · Public · internals redacted · Aug 2026
[ 24 / 25 ]
Slide 25 — The Ask & CAPEX Ceiling
⚡ EDHTA · paper v3
The ask
Stage 1 pilots now · Stage 2 owned mind CAPEX (planning ceiling · upper +5%)
SAST pilot
DAST pilot
Design partners
Authorised code-review assist
Authorised dynamic-test assist
Industrial / air-gap talks
Stage 2 capital — owned Ghostmaster v11 headless mind (planning ceiling)
Mind: dual 96 GB class GPUs · headless Linux appliance (owned only — not cloud
rent)
Upper BOM bound × 5% contingency.
Operator: Linux laptop plane (System76 Lemur Pro 14 class) · does not host the
Soak one GPU first. Re-quote before purchase.
dual-GPU mind
Mind
Laptop
System
High + import
$40k
₹39L
$3.2k
₹3.0L
$43k
₹42L
$61k
₹59L
Soumabha Majumdar · Founder & CEO · prior corporate developer + infosec engineer
MSc Digital Forensics (NFSU) · BTech · CEH · eWPTX · soumabha@noctrodex-hq.com · noctrodex-hq.com
Public · EDHTA paper v3 company IP · planning ceilings not firm quotes · Stage 1 + Stage 2 · Aug 2026
