Hysteresis logo
Hysteresis · Pvt. Ltd.
Portfolio · 2026 Pune / Bangalore

Engineer of unlikely
systems. Nine production builds across AI, infra, IoT, mobile, and commerce.

I build software that has to actually work — autonomous AI agents, enterprise-grade APIs, predictive maintenance pipelines, and the occasional multi-tenant smart-home stack. Founder–engineer, ex-cloud architect, currently building Eval8.ai.

GKV
⌁ FOUNDER · CEO
AI · INFRA · IoT
3+ YRS AWS
9 LIVE BUILDS
NVIDIA Partner Naukri / DoSelect Torry Harris Amdocs Hindustan Unilever Smallest.ai Eval8.ai Vishq.io Habitica NaturesDeli NVIDIA Partner Naukri / DoSelect Torry Harris Amdocs Hindustan Unilever Smallest.ai Eval8.ai Vishq.io Habitica NaturesDeli
⏺ Verified by 72 customers

People trust
what we ship.

Independently reviewed on Trustpilot — across enterprise training engagements, product builds, and freelance work. The reviews aren't paid, prompted, or polished. They're from people who actually went through the work.

Hysteresis Pvt Ltd · Verified
4.4
TrustScore · 72 reviews
"Gokul's training on every topic is clear cut and crisp. The training was fully packed and made us curious on cloud."
— Verified · Torry Harris cohort
"The way he taught us was brilliant. Initially fast, but I kept up — and learned more than I thought possible."
— Verified · THBS engineer
65% 5★ 32% 4★ 3% 3★
Read all 72 reviews →
⏺ Strategic Partnership · 2026

Hysteresis is now an
official NVIDIA partner.

Joining NVIDIA's partner ecosystem unlocks direct GPU compute access, early CUDA & cuDNN tooling, and joint go-to-market on enterprise AI infrastructure deployments. For Eval8.ai, that means on-prem inference at scale — sovereign LLMs running on NVIDIA hardware inside customer perimeters.

⚡ NVIDIA Inception ⌁ CUDA Toolkit Access ◆ DGX-Class Compute ⚙ TensorRT · Triton
Verified Partner
NVIDIA Partner
Hysteresis Pvt. Ltd. Approved · Certified · Active
Selected Engineering Work · 01 ⟶ 09

Things I've shipped, broken, and shipped again.

01 / FLAGSHIP ⏺ LIVE · early revenue

Eval8.ai — Autonomous Learning & Hiring Engine

End-to-end AI evaluation · Founder & CEO · Hysteresis Pvt. Ltd.
⟁ ALCTS
Curriculum auto-generated
from 47 JDs · gap-mapped
to 1,240 sub-skills · live
◉ SAII Interviewer
Sovereign · on-prem
Whisper STT → LLM → Kokoro
142ms turn latency
⌁ Behavioral Integrity
CDP screen capture
Qwen-VL gaze + tab events
score: 0.91 confidence
⚡ Prompt-to-Hire
"find me 5 backend devs
who know Rust + k8s
in Pune under ₹18LPA" → ✓

A vertically integrated platform replacing the assessment, training, and proctoring stack for engineering colleges and recruiters. Built on a local-first AI architecture that hits ~90% gross margins at ~₹12,999/student — a number nobody in this category has touched.

Hyper-innovative bits

  • Autonomous Learning & Curriculum Tuning System (ALCTS) — ingests live job descriptions, reverse-maps them to skill graphs, and auto-generates per-student curriculum that mutates as the market shifts.
  • Sovereign AI Interviewer (SAII) — runs entirely on-prem on Apple Silicon: Whisper STT → local LLM → Kokoro TTS. Zero data leaves the college network. Patent-pending.
  • Behavioral Integrity Scoring — Chrome DevTools Protocol screen capture + Qwen vision model fuses gaze, tab-switching, and keystroke biometrics into a single integrity score. No proctor needed.
  • Prompt-to-Hire Engine — recruiters describe a candidate in plain English; the engine retrieves, ranks, and books interviews end-to-end.
  • AI Desktop Learning Agent "Tux" — a personality-driven agent that drives VSCode, browser, and terminal to teach by demonstration. Four mood modes.
ElectronNode 20 WhisperKokoro TTS Qwen-VLRough.js CDPFastAPI PostgresAWS Apple Silicon (Metal)
02 ✓ SHIPPED · production

DoSelect API Gateway

Naukri.com · Candidate parsing & assessment routing

Built the core API gateway powering candidate-resume parsing and assessment dispatch for DoSelect (a Naukri-owned product). The kind of system where a 200ms regression is somebody's bad Monday morning.

Hyper-innovative bits

  • Schema-aware throttling — rate limits adapt per endpoint based on payload entropy, not just request count.
  • JWT-with-rotating-claims — tokens carry per-tenant feature flags that flip without re-issuing.
  • Multi-format resume parser — PDF, DOCX, image-based; falls through OCR with confidence scoring.
Node.jsExpress RedisJWT AWS LambdaAPI Gateway CloudWatch
03 ✓ LIVE

NaturesDeli

Grocery e-commerce platform

End-to-end grocery commerce stack — catalog, cart, payments, last-mile dispatch. Built for cold-chain SKUs with route-optimized delivery windows.

Hyper-innovative bits

  • Cold-chain aware fulfilment — SKUs tagged with thermal envelopes; dispatcher refuses incompatible co-loads.
  • Inventory de-duplication across third-party suppliers.
Next.jsStripe PostgresAWS S3
04 ◐ IN PROGRESS

Vishq.io

Smart-home automation OS
5 nodes · MQTT · 24ms avg

A protocol-agnostic smart-home control plane. One UI, every device.

Hyper-innovative bits

  • Mesh-first — devices coordinate locally even if internet drops.
  • Intent automation — "wind down for the night" resolves to a learned action graph.
MQTTZigbee2MQTT React NativeRust (edge)
05 ✓ DELIVERED

Smallest.ai TTS

Internal freelancer · Protocol & param testing
SAMPLE_RATE: 24000Hz LATENCY_P99: 84ms ✓ 12 voices · 9 langs · benched

Tested and tuned TTS protocols and parameters across voices and languages for one of India's fastest TTS startups.

Hyper-innovative bits

  • Latency-perceptual tradeoff harness — automated benchmarking against MOS scores.
  • Streaming-first first-token-out budgeting under 100ms.
PythonWebSockets PyTorchgRPC
06 ✓ DELIVERED

Torry Harris × Azure Entra

Custom authentication protocols on Microsoft Entra ID
CLIENT ENTRA ID + CAE RESOURCE Conditional Access · Token Bind · MFA PKCE + DPoP · token leak ↓ 100%

Designed custom auth flows on Azure Entra ID — Conditional Access policies, DPoP-bound tokens, and a custom claims pipeline that reduced token-leak surface materially.

Hyper-innovative bits

  • DPoP-bound tokens — proof-of-possession tied to client key; replay-attack resistant.
  • Continuous Access Evaluation with custom risk signals from device posture.
Azure EntraOAuth 2.1 DPoP.NET Conditional Access
07 ✓ DELIVERED

Amdocs · Prometheus & Grafana

Observability stack for telco-grade workloads
node_cpu_usage{quantile="0.99"} 15m · 1h · 6h · 24h · 7d

Stood up the metrics + alerting backbone for high-volume telecom workloads — custom exporters, alert routing, and SLO dashboards that ops actually opens.

Hyper-innovative bits

  • SLO-based alerting — burn-rate alerts replace static thresholds.
  • Multi-tenant Grafana with per-team RBAC and folder-scoped dashboards.
PrometheusGrafana AlertmanagerPromQL Kubernetes
08 ✓ DELIVERED · workshop included

Hindustan Unilever · Predictive Maintenance

ML pipeline for industrial uptime + capability transfer workshop
SENSORS INGEST FEATURES MODEL ALERT + DIGEST vibration temp current Kafka 10K/s FFT rolling stats XGBoost RUL Slack · SAP PM-order

Built a predictive maintenance solution for HUL's manufacturing lines — sensor ingest through XGBoost-based remaining-useful-life modeling, with automated work-order creation back into SAP. Then ran a workshop to hand it over to the in-house team.

Hyper-innovative bits

  • FFT-derived features — vibration spectra collapsed into a 32-dim feature vector that XGBoost handles cleanly.
  • RUL (Remaining Useful Life) regression instead of binary classification — operators get a horizon, not just a yes/no.
  • SAP-PM auto-ticketing — alerts don't end in Slack; they create the maintenance work order.
  • Capability transfer workshop — full hand-off so HUL's team owns it, not us.
PythonXGBoost KafkaAirflow SAP PMAzure ML Grafana
09 ✓ SHIPPED · Android native

Habitica — Gamified Productivity

Native Android · Android Studio · XML rendering protocols
⚔ QUEST LOG
HP 72 / 100
XP 45 / 100
Morning workout
Code review · 2hr
Read 30 pages
Hydrate · 8 cups
+12 XP today
⟁ XML LAYOUTS
<LinearLayout
  android:orientation=
  "vertical">
  <CardView
    app:cardElevation=
    "4dp">
    <ProgressBar
      style="?xpBar"/>
  </CardView>
</LinearLayout>
28 layouts

A gamified personal-productivity Android app — built natively in Android Studio with Java/Kotlin and pure XML-based view rendering. Every habit, daily, and to-do becomes a quest; consistency earns XP and gold, slipping costs HP. Built from the ground up with custom XML layouts (no Jetpack Compose), demonstrating mastery of Android's classic rendering pipeline.

Hyper-innovative bits

  • Pure XML rendering pipeline — every screen built with native Android view hierarchies (LinearLayout, ConstraintLayout, RecyclerView, CardView), with custom drawable XMLs for state-driven animations. No Compose, no shortcuts.
  • Dual-currency RPG mechanics — HP/XP/Gold economy where positive habits add XP and missed dailies subtract HP, creating real behavioural feedback loops.
  • Custom view components — built reusable XP progress bars, quest cards, and avatar widgets as <custom-view> XML tags with Kotlin-backed render protocols.
  • Selector & state-list drawables — task buttons morph through pressed/checked/levelled-up states using XML state machines, no third-party animation libs.
  • Material Design adherence — every dimension, padding, and elevation traced back to Material spec via styles.xml + themes.xml.
  • Offline-first room database — SQLite via Room, so quests persist even without connectivity. Sync layer queues actions for later.
Android StudioKotlin JavaXML Layouts ConstraintLayoutRecyclerView Room DBSQLite Material DesignDrawable XML

Meet the person behind it.

Gokul Kumar Vishwanathan
GV
GOKUL KUMAR
VISHWANATHAN
FOUNDER · CEO
📸 save photo as "gokul.jpg" beside this file
Founder & CEO
Gokul Kumar
Vishwanathan
Founder & CEO · Hysteresis Pvt. Ltd.
AWS Certified Azure AI Engineer 5+ yrs DevOps Patent Holder
"

It's never too late to start learning. We drive continuous transformation — embedding technology deeply into every organisation we touch.

— Gokul Kumar Vishwanathan, Founder & CEO

Gokul Kumar Vishwanathan is a serial entrepreneur, corporate trainer, patent holder, and published researcher based in Pune, India. He holds a B.Tech in Mechanical & Robotics Engineering from MIT Academy of Engineering, Alandi — and has built an extraordinary bridge from deep engineering roots to enterprise technology leadership.

As a corporate trainer, Gokul has personally upskilled workforces at 15+ major MNCs — including Torry Harris Business Solutions, Deloitte, Hindustan Unilever, Persistent, and Amdocs — earning a reputation for fast-paced, hands-on pedagogy that transforms careers, not just CVs. His current focus spans building Eval8.ai, scaling Vishq.io, and driving enterprise cloud and AI transformation across India.

🏅
AWS & Azure Certified
Azure AI Engineering · Data Engineering · AWS Glue · 5+ yrs Cloud DevOps
📜
Patent Holder
Granted patent — recognised technology innovator
🎓
Published Researcher
Peer-reviewed research paper in technology
🎤
Conference Speaker
aiDevCon India 2025 · DevSecOps with AI
🏢
Ex Tata Consultancy Services
Systems Engineer · TCS · tier-1 enterprise
🚀
Multi-Venture Director
Director · Vedhkart Pvt. Ltd. · India ecosystem
📱
NaturesDeli on PlayStore
Built & launched a live Android application
💬
Built neXTerchat SaaS
Social platform for tech enthusiasts
Tooling I reach for

A working toolbox, not a buzzword bingo card.

⌁ AI / ML / Deep Learning
Transformers · Attention Mechanisms · Self-Attention · Multi-Head Attention
PyTorch · TensorFlow · JAX · Hugging Face Transformers
ReLU · GELU · LeakyReLU · Softmax · LayerNorm · BatchNorm
Backpropagation · Gradient Descent · Adam · AdamW · RMSProp
RAG · Vector DBs (FAISS, Pinecone, Chroma) · Embeddings · Cosine Similarity
Fine-tuning · LoRA · QLoRA · PEFT · Quantization (INT8 / GPTQ)
CNN · RNN · LSTM · GRU · ResNet · BERT · GPT-class decoders
XGBoost · LightGBM · Random Forest · SHAP · Feature Engineering
◆ Cloud / Infra / Data Engineering
AWS (EC2, Lambda, S3, ECS, EKS, RDS, SQS, SNS, CloudFront)
AWS Glue · Athena · Redshift · Kinesis · EMR · Step Functions · Data Lakes
Azure AI Engineering · Azure ML · Synapse · Data Factory · Cognitive Services
Azure Entra ID · Conditional Access · DPoP · OAuth 2.1 · OIDC
Kubernetes · Helm · Istio · Service Mesh · Horizontal Pod Autoscaling
Terraform · Pulumi · CloudFormation · Ansible · Infrastructure-as-Code
Prometheus · Grafana · PromQL · Loki · Alertmanager · OpenTelemetry
Docker · containerd · CI/CD · GitHub Actions · ArgoCD · GitOps
⟁ App / Web / Backend
Next.js (App Router, RSC) · React 18 · TypeScript · Vite
Node.js · Express · NestJS · FastAPI · gRPC · GraphQL · REST
PostgreSQL · MongoDB · Redis · Elasticsearch · Kafka · RabbitMQ
Electron · React Native · Tailwind · WebSockets · Server-Sent Events
Auth · JWT · OAuth 2.1 · OIDC · PKCE · SAML · session security
Microservices · Event-driven architecture · CQRS · DDD
◉ Edge / IoT / Native
MQTT · Zigbee · Z-Wave · BLE · Matter / Thread
Rust (embedded · no_std) · C++ · ESP-IDF · Arduino · Raspberry Pi
Android Studio · Kotlin · Java · XML Layouts · Room · Jetpack
Apple Silicon · Metal Performance Shaders · CoreML
Chrome DevTools Protocol · Puppeteer · Playwright · Headless automation
FFT · DSP · sensor fusion · Kalman filters · time-series forecasting

Let's build
something
unreasonable.

Whether it's an autonomous system, a brutally fast API, or that one piece of infra you've been putting off — I'm available for select engagements through Hysteresis.

Gokul Kumar Vishwanathan
Founder · CEO · Engineer
Hysteresis Pvt. Ltd. Pune · Bangalore · India
Eval8.ai