QGAI · Quantum Generative AI Lab · Pune, India
Teaching machines toimagine in superposition, reason over entanglement, sample the impossible, compute beyond classical
I'm Prakash Gore. I build hybrid quantum-classical systems that make generative AI more data-efficient, more expressive and more trustworthy — and I engineer them to run on the hardware that exists today, not the hardware we're promised for 2035.
- Active research tracks
- 6
- Open-source research notebooks
- 32
- Qubits in simulated circuits(sim.)
- 40+
- Years engineering at scale
- 18+
About · The founder
A builder who reads the papers — and a researcher who ships.
Most quantum research never leaves the notebook. Most enterprise AI never questions its own assumptions. QGAI exists in the gap between the two.
My career began in large-scale software engineering — distributed systems, cloud platforms and the unglamorous discipline of making things work at 3 a.m. Over the years I moved from writing services to architecting AI solutions for enterprises: data platforms, ML pipelines, and, more recently, LLM and retrieval systems serving real users with real compliance constraints.
Quantum computing found me the way it finds many engineers — through a single question I couldn't let go of: if generative models are fundamentally about learning probability distributions, why are we ignoring the only computers that natively are probability distributions?
QGAI is my answer: a research lab that treats quantum computers as specialised accelerators inside AI systems, the way GPUs became accelerators for graphics and then for everything. Not hype. Not a moonshot slide. A methodical programme of experiments, benchmarks and open engineering.
Vision
“By 2030, every serious AI platform will treat the QPU as a first-class accelerator alongside the CPU and GPU. QGAI's mission is to make that integration rigorous, measurable and open — and to make sure India helps write the playbook.”
— Prakash Gore
01
Advantage must be measured
Every claim is benchmarked against the strongest classical baseline we can build — including tensor networks and GPU simulators.
02
Hybrid by default
Quantum circuits do what only they can do. Classical systems do the rest. The architecture is the innovation.
03
Open over opaque
Methods, negative results and reproducible notebooks are published. Science compounds when it is shared.
04
Engineered for noise
Designs assume NISQ-era hardware and early fault tolerance — error-aware, shallow and mitigation-ready.
Research areas
Six tracks. One hypothesis.
Generative AI is the art of learning and sampling probability distributions. Quantum computers are probability distributions. Each track tests a different consequence of that idea — with classical baselines that are hard to beat.
Innovation lab
The lab notebook, open.
Every experiment starts with a falsifiable hypothesis and ends with a result — positive or not. Here's what's on the bench right now.
Live experiment log
6 active
EXP-014
Quantum RAG
QUBO context packing for multi-hop RAG
H₁: Jointly optimising relevance and redundancy beats greedy MMR on multi-hop questions at equal token budget.
Validating72%updated 2026-10
EXP-011
Synthetic Data
Born-machine priors for low-data tabular synthesis
H₁: A 12–20 qubit QCBM prior improves rare-category coverage versus a matched classical prior.
Prototyping48%updated 2026-09
EXP-009
Quantum + HPC
Latency budget of cloud QPUs in variational loops
H₁: Queue and transfer overheads, not gate time, dominate wall-clock for VQE-style loops — and are schedulable.
Validating81%updated 2026-10
EXP-007
Knowledge Graphs
Density-matrix embeddings for polysemous entities
H₁: Mixed-state embeddings outperform point vectors on link prediction for multi-sense entities.
Exploring26%updated 2026-08
EXP-005
Quantum-AI Convergence
Verifier-in-the-loop circuit generation agent
H₁: An LLM agent with simulator feedback produces shallower circuits than one-shot generation.
Prototyping55%updated 2026-10
EXP-003
Quantum GenAI
Barren-plateau map of shallow generative ansätze
H₁: Problem-inspired, local-cost ansätze retain trainable gradients up to ~30 qubits in simulation.
Piloting90%updated 2026-07
Open Benchmarks
A reproducible suite comparing quantum and classical generative models on equal footing — including negative results.
Quantum Readiness Programme
Working sessions that help engineering teams identify realistic quantum use-cases and post-quantum migration needs.
Bharat Quantum Builders
Mentoring and open notebooks for Indian students and engineers entering quantum software, aligned with the National Quantum Mission's talent goals.
Technology exploration · on the radar
- Post-quantum cryptography migration patterns for AI platforms
- Tensor networks as the honest classical baseline
- Quantum reservoir computing for time-series
- Photonic and neutral-atom hardware trade-offs
- Error-mitigation cost models (ZNE, PEC, PEA)
- qLDPC codes and what they mean for software timelines
Projects showcase
From hypothesis to running system.
Research becomes real when it has an architecture, an API and an honest metric. Each project below has all three.
6 projects shown
- QuantumGenAIPrototyping
Q-Sampler
A hybrid generative engine that plugs a quantum circuit Born machine in as a learnable prior for classical decoders.
- Rare-mode coveragetarget
- +15–25%
- Qubits (simulated)benchmark
- 8–24
- Baselines tracked
- 4
- PennyLane
- PyTorch
- CUDA-Q
- Weights & Biases
- Docker
- HybridGenAIValidating
EntangledRAG
A retrieval layer that formulates context selection as a QUBO and solves it with QAOA, annealing or a classical fallback.
- Context redundancytarget
- −30%
- Candidate poolbenchmark
- ≤64
- Solver back-ends
- 3
- Qiskit
- D-Wave Ocean
- pgvector
- LangGraph
- FastAPI
- Next.js
- HybridPlatformValidating
QHPC Orchestrator
A scheduler that co-locates GPU simulation and cloud-QPU jobs, routing each circuit by cost, fidelity and queue time.
- Idle queue timetarget
- −40%
- Providers abstracted
- 3
- Shot budget savedtarget
- 20%+
- Kubernetes
- Ray
- Qiskit Runtime
- Amazon Braket
- Prometheus
- Go
- QuantumPlatformPrototyping
QSynth
A privacy-first synthetic tabular data toolkit with quantum copula generators and built-in utility and privacy reports.
- Tail fidelitytarget
- +20%
- Privacy budget εtarget
- ≤ 3
- Report metrics
- 12
- Qiskit
- SDV
- Opacus
- Polars
- Streamlit
- GenAIQuantumPrototyping
Qircuit Copilot
An agentic assistant that drafts, simulates and optimises quantum circuits — every suggestion is verified before it is shown.
- Circuit depthtarget
- −25%
- Verified outputs
- 100%
- Task benchmarktarget
- 150 tasks
- Claude / LLM APIs
- Qiskit
- Aer
- Tool-use agents
- TypeScript
- QuantumGenAIExploring
Hilbert KG
Quantum-inspired knowledge graph embeddings that model ambiguity as mixed states, feeding a GraphRAG pipeline.
- Link-prediction MRRtarget
- +8%
- Entities (pilot)target
- 50k
- Embedding families
- 3
- PyKEEN
- PyTorch (complex)
- Neo4j
- pgvector
- LlamaIndex
Research impact dashboard
Progress you can audit.
Output metrics for the open research programme, refreshed monthly.
32
Notebooks open-sourced
Reproducible, with seeds and baselines
≥ 3
Classical baselines per study
Including tensor-network simulation
9
Negative results published
Because null results save others months
120+
Engineers mentored
Through workshops and open sessions
Simulated circuit runs per month (thousands)
May–Oct 2026 · 199k total · +206% over the period
View as table
| Month | Runs (k) |
|---|---|
| May 2026 | 18 |
| Jun 2026 | 24 |
| Jul 2026 | 31 |
| Aug 2026 | 29 |
| Sep 2026 | 42 |
| Oct 2026 | 55 |
Publications & insights
Thinking in public.
Articles, whitepapers, research summaries and technical deep-dives — written to be useful to researchers and engineering leaders alike.
Technology stack
An ecosystem, not a toolbox.
Hybrid research lives at the seams between ecosystems. Select any technology to trace how it connects across quantum, AI and cloud.
Quantum
Cloud
AI
How to read this map
Each column is an ecosystem; each curve is a working integration used in at least one QGAI project. Hover, tap or tab through nodes to trace a dependency chain.
Future roadmap · 2027 → 2030
The decade the QPU joins the stack.
A forecast is a hypothesis with a deadline. Here is ours — what we expect the industry to do, and what QGAI commits to building.
- 2027
Utility in the loop
Industry prediction
Early fault-tolerant demonstrations with tens of logical qubits become routine. Hybrid quantum-HPC centres move from pilots to shared research infrastructure, and India's National Quantum Mission platforms reach the intermediate scale.
QGAI research direction
Publish the QGAI open benchmark for quantum generative models. Ship EntangledRAG and QHPC Orchestrator as open-source reference implementations.
- 2028
The QPU becomes an accelerator
Industry prediction
Major AI and cloud platforms expose QPUs through the same schedulers as GPUs. First credible, narrowly-scoped advantage results in chemistry sampling and optimisation sub-routines.
QGAI research direction
Industry pilots of quantum synthetic data in finance and healthcare. Release a quantum-programming benchmark for code models.
- 2029
Logical-qubit software
Industry prediction
Software shifts from physical-gate thinking to logical-qubit abstractions. Error-correction decoding — often AI-driven — becomes a core part of the stack. Post-quantum cryptography migration deadlines force enterprise action.
QGAI research direction
Port QGAI models to logical-qubit back-ends. Launch a quantum-AI convergence fellowship with Indian universities.
- 2030
Quantum-native AI
Industry prediction
Hundreds of logical qubits make genuinely quantum generative models — not just quantum-inspired ones — testable at meaningful scale. AI and quantum co-design each other.
QGAI research direction
Quantum-native foundation components: priors, samplers and reasoning modules deployed inside production AI systems.
Calibrated predictions
Stated confidence that each holds by end of 2030 — we'll score ourselves publicly.
The first commercially meaningful quantum advantage will be in sampling, not search.
70%Most 'quantum AI' value before 2030 will come from quantum-inspired classical algorithms.
80%AI-driven decoders will be standard in fault-tolerant control stacks.
65%India will host at least one globally-relevant quantum software company by 2030.
60%
Contact & collaboration
Build the quantum-AI stack with us.
Universities, research labs, enterprises and event organisers — tell us what you're working on. Every message is read by Prakash personally.