Co-Founder & CTO @ Dehurdle | IIT Delhi
Building intelligence infrastructure for human performance.
I believe the current "Generative AI" era is a dead end for true enterprise intelligence. Predicting the next text token does not equate to understanding reality. True behavioral intelligence requires mapping the underlying physics of a system—moving away from autoregressive generation and towards predicting deterministic state transitions (
At Dehurdle, we don't build HR chatbots. We build mathematical state machines.
My daily engineering work sits at the intersection of behavioral science, edge-compute, and cryptographic privacy:
- Zero-Payload Edge Ingestion: Bypassing the AI "Payload Trap." We process non-lexical metadata at the edge to measure structural friction and organizational latency, without ever reading a private employee message. CISO-compliant from Day 1.
- Deterministic Temporal State Machines: Replacing hallucination-prone LLM RAG pipelines with strict temporal state vectors that track behavioral decay, momentum, and phase transitions over time.
- Tiered Edge-to-Cloud Compute: Decoupling cheap signal ingestion (edge SLMs) from heavy generative reasoning (cloud LLMs) to guarantee 85%+ software margins at scale.
If you are looking at my commit graph below, you will see a massive amount of velocity (3,600+ commits) but no public code.
Because Dehurdle deploys infrastructure into highly regulated Fortune 500s, global conglomerates, and BFSI environments, 100% of my core engineering work—our edge classifiers, coupling matrices, and proprietary inference layers—lives in strictly private, SOC2 and DPDP-compliant enterprise repositories.
- Twitter: Vishal
- LinkedIn: in/visha1v (Note: adjust if your URL handle is slightly different)
- Company: dehurdle.com
- Personal Website: vishalvermalabs.com