90%
Krrish Choudhary
Founder & CEO
- ·Wrote every line of VCI from first principles.
- ·Papers: The Artificial Brain · StreamRAG v2 (IJTRP, Feb 2026).
- ·Ex-Incredible (AI platform, 100+ integrations) · Ex-GoQuant Miami.
- ·LNMIIT Jaipur · Class of 2027.
PRE-SEED · 2026
Verace-AGI
Founder & CEO
Krrish Choudhary
krrish@verace.in
verace.in · Jaipur, India
Architectural Limitation
The transformer architecture lacks three foundational primitives required for trustworthy AI in regulated environments.
01Absent
Memory
No persistent storage of context across sessions or interactions.
Consequence
Every conversation starts from zero. No customer history, no case context, no compounding institutional knowledge.
Current workaround
RAG + vector DBs
External retrieval is not internal recall. Bolt-on memory cannot reason across what it just retrieved.
02Absent
Self-Awareness
No internal confidence signal at the point of inference.
Consequence
Hallucinates with full confidence. Cannot flag its own uncertainty. Cannot say “I don’t know.”
Current workaround
Logprob-based proxies
Token probabilities are not calibrated truth signals. Regulated domains cannot underwrite proxy uncertainty.
03Absent
Local Learning
No mechanism to update knowledge without full retraining.
Consequence
One new fact requires a full fine-tune cycle. Each update costs millions and risks catastrophic forgetting.
Current workaround
Fine-tuning + prompt engineering
Both are external mitigations layered on a frozen substrate. Neither produces per-fact, in-place updates.
Market Implication
These are architectural constraints, not accuracy ceilings. External mitigation through retrieval, fine-tuning, or extended context cannot substitute for the missing primitives. Industries that require deterministic memory, calibrated uncertainty, and continuous learning remain structurally underserved by the current generation of models.
Inflection
The architectural replacement market opened in March 2026 with a ₹8,600 Cr seed round. JEPA is one approach. Cortical columns are the other.
01Past
2017
Substrate published
Transformer architecture released as an open paper with no defensive IP. Every LLM since is a refinement of the same substrate.
02Past
2022
Critique enters mainstream
ChatGPT ships in November. Six months earlier, LeCun publishes the JEPA position paper. Capability and architectural critique reach the field in the same year.
03Competitor · funded
2026 Mar
AMI Labs validates
AMI Labs (LeCun) raises ~₹8,600 Cr seed at ~₹38,000 Cr post-money to build JEPA world models. Bezos Expeditions, Nvidia, Samsung back the architectural thesis.
04Verace · ready
2026 Now
Verace competes
Cortical column network shipped. Seven pillars implemented end to end. The architecture-layer race is on. JEPA one approach, cortical columns the other.
Implication
The next moat is at the architecture layer, not the parameter layer. AMI’s raise priced the category. Verace is the differentiated alternative: instead of world-model latent prediction, a modular cortical column network with persistent memory, calibrated uncertainty, and local learning. Two architectures, one window.
Architecture
Verace Cortex Intelligence is a modular cortical column network. The third defensible architecture in the post-transformer era.
Three missing primitives, three VCI mechanisms
01Primitive
Memory
→
Persistent Episodic Memory.
One-shot encoding into writable structures. Recall across sessions. Memory is data, not weights. Add, retrieve, or revise without retraining.
02Primitive
Self-Awareness
→
Per-Step Confidence Tracking.
Inference-time uncertainty signal at every column. Deliberation triggered when confidence falls below threshold. The system reports what it does not know.
03Primitive
Local Learning
→
Per-Column Update Protocol.
Each column corrects from the input it just saw. No global gradient graph. Update cost is constant in model size. Learning is per-fact, not per-checkpoint.
Architectural primitive · post-transformer landscape
Transformer.Incumbent
Token attention
Refines next-token prediction at scale. No persistent memory, no confidence signal, no per-fact learning.
JEPA.Funded competitor
Latent world models
Predicts in embedding space rather than token space. Same training-then-freeze paradigm. AMI Labs raised ₹8,600 Cr in 2026 to build it.
VCI.Verace
Cortical column network
Specialized columns with persistent memory, calibrated uncertainty, and per-column local learning. Biological primitives.
Implication
The post-transformer era has two viable primitives. JEPA pursues latent prediction; VCI pursues biological cortical mechanisms. Both are architecturally defensible and uncorrelated. The market priced one of them at ₹8,600 Cr. Verace is the second.
Market
Two flagship verticals address ~₹1,00,000 Cr Indian TAM by 2030. India first deployment. Global expansion to follow.
TAM by vertical · 2030
₹ Cr · log-equivalent scale not used
Bottom-up sizing
#Vertical · GeoAddressable unitsPricing modelSOM · 3-yr commitment
01Legal.India1.4M+ registered Indian advocates · 50K+ firmsPer-seat SaaS + enterprise licensing500 Indian firms by 2028
02Healthcare.India90,000+ registered Indian clinicsPer-clinician monthly SaaS200 clinics + 5 hospital systems by 2029
Adjacent · 2028–2029
03FinanceCompliance + risk memo retention· 2028
04EducationTutors with year-long student memory· 2028 H2
05EmbeddedOn-device cortical models· 2029
Implication
India first by design. Deterministic memory for Indian legal precedent and case law, calibrated uncertainty for Indian clinical reasoning. ~₹1,00,000 Cr combined Indian TAM by 2030. Global expansion follows India validation.
Product
Seven products derive from one substrate. Sequenced launch from 2027 H1 through 2029.
Launch sequence · 2027 H1 → 2029
4 primary · 3 adjacent · 7 total
Product portfolio
#ProductPricing modelTarget customerLaunchStatus
01VCI API.Per-tokenDevelopers · AI product teams2027 H1Primary
02VCI Enterprise.Annual licensingBanks · defense · sovereign cloud2027 H2Primary
03VCI Legal.Per-seat SaaSIndian law firms · 50K+2027 H2Primary
04VCI Health.Per-clinician SaaSIndian clinics + hospitals2028Primary
05VCI Finance.Per-seat SaaSCompliance + risk teams2028Adjacent
06VCI Edu.Per-student SaaSIndian K-12 + higher ed2028 H2Adjacent
07VCI Embedded.License + SDKHardware OEMs · IoT · robotics2029Adjacent
Implication
Platform anchored by vertical ownership. Same substrate underwrites every product, so each new vertical ships at marginal incremental cost. Pricing models span per-token API, per-seat SaaS, and annual enterprise licensing — the revenue stack on the Business Model slide derives from this portfolio.
Traction
Architecture complete. Training in flight. Substrate validated by 84 inventions and 7 patents filed since March 2026.
7
Patents pending
Filings on every core architectural claim that separates VCI from incumbents.
84
Inventions catalogued
Mechanism-level claims, each implemented and live in the running system.
7 / 7
Pillars of intelligence
All seven cognitive primitives implemented end-to-end in the cortical column network.
Milestone timeline · Mar 2026 → 2027 H1
1 complete · 1 in flight · 2 planned
Implication
Substrate built in ~120 days from company founding. ARR is not the right traction metric for a deep-tech architecture company at pre-seed; IP filings, architectural completeness, and training progress are. The ARR conversation begins at API launch in 2027 H1.
Business Model
Five revenue streams cover the full pricing spectrum — from per-token API to annual enterprise licensing.
Sequence of activation · 2027 H1 → 2029+
5 streams · launched one after another · each unlocks the next
Revenue stream portfolio
#StreamPricing modelTargetACV bandLaunch
01Usage-based API.Per-tokenDevelopers · AI product teamsLow ACV · High volume2027 H1
02Legal SaaS.Per-seat · monthlyIndian law firms · 50K+Med ACV · Med volume2027 H2
03Healthcare SaaS.Per-clinician · monthlyIndian clinics + hospitalsMed ACV · Med volume2028
04Enterprise licensing.Annual · on-premBanks · defense · sovereign cloudVery high ACV · Low volume2028 H2
05Custom integrations.Project-basedBespoke VCI deploymentsHigh ACV · Very low volume2029+
Cost Structure
Linear
Local learning per column means inference cost scales linearly with volume. No per-update retraining tax.
Gross Margin
Structurally favorable
No global-gradient retraining cost. Margin profile better than transformer LLM incumbents at every contract size.
LTV / CAC
In progress
Target ratios and comparable benchmarks modeled in the appendix. Discussed in person.
Implication
Sequenced, not scattered. API first in 2027 H1 to validate the substrate; SaaS verticals layer on through 2028; enterprise + custom anchor 2028 H2 onward. Five streams across the full ACV spectrum — same VCI substrate underwrites every one of them, so incremental margin cost per new stream is near zero.
Moat
Scaling a transformer is not a moat. Owning a different architectural substrate is.
Architecture × seven-pillar coverage
● full · ◐ partial · ○ none
7
Patents pending
Filings on every core architectural claim that separates VCI from incumbents.
84
Inventions catalogued
Mechanism-level claims, each implemented and live in the running system.
7 / 7
Pillar coverage
Only post-transformer architecture covering all seven cognitive primitives.
Implication
The moat is at the substrate layer, not the weights layer. A competitor cannot reach VCI by training a larger transformer or by scaling JEPA — they would have to rebuild a different cortical substrate from scratch, and the architectural primitives are what we have been filing.
Go-to-Market
Three sales motions in sequence: bottom-up API, land-and-expand SaaS, top-down enterprise.
GTM sequence · 2027 H1 → 2028 H2
3 sales motions · each unlocks the next
Sales motion spectrum
Implication
Each motion validates the next. API proves the substrate. Legal design partners prove the workflow. Enterprise contracts prove the ACV. The same VCI substrate underwrites all three; only the sales motion changes per channel.
Team
Three people. 100% founder-held. Founder wrote every line of VCI from first principles.
5%
Gautam Mittal
COO
- ·Operations, product, and early go-to-market.
- ·Runs partnerships and investor-facing motion.
5%
Tanvi Kandoi
Jr. Engineer
- ·Performance optimizations across VCI.
- ·Kernel-level tuning and runtime profiling.
Cap table · 100% founder-held
pre-investment · ESOP + investor allocation finalized at closing
Implication
Three people. One substrate. Founder wrote every line of VCI from first principles. Hiring plan + ESOP pool discussed in person.
Pre-seed · Open
₹5 Cr for 5%
₹100 Cr post-money valuation · clean cap table · founder team retains 95%
₹5 Cr unlocks·
NOWTraining at scale
─→2027 H1VCI API live
─→2027 EOYFirst ₹50 Cr
Use of funds· ₹5 Cr deployed
Signed
Krrish Choudhary
Founder & CEO · Verace AGI