Skip to content

Dibyaprakash Pradhan

I build AI foundation architectures and models at AntEngage. Designed from first principles rather than derived from the transformer lineage — built for efficiency and capability the current generation is architecturally unable to reach, and meant to hold for the next two decades.

At RoboDIB I build pcbeditor.com — describe a circuit in plain English and get a verified schematic and a complete, fabricable board. It is AI-native the whole way down: schematic, circuit, footprints, netlist and trace routing are all generated, then AI design-rule checking and copper pour, with 3D stack-up and board views. Free, in the browser, exports to KiCad and Gerber.

Currently — foundation model architecture · circuit synthesis · Bengaluru, India

§01Building
  • Founder & CTO

    AntEngage

    The company I lead as Founder & CTO, and where the foundation model work happens. We are building model architectures from first principles rather than deriving them from the transformer lineage — the thesis being that the efficiency and capability ceiling of the current generation is an artefact of its architecture, not of what is possible. AntEngage also ships conversational AI for engagement, which is where much of my voice and real-time systems work comes from.

    AI foundation models · antengage.com

  • Founderat RoboDIB

    PCBEditor

    An AI-native circuit design tool. Describe a circuit in plain English and it generates the schematic, the circuit itself, component footprints, the netlist and the routed traces — then runs AI design-rule checking, handles copper pour, and renders the layer stack and the finished board in 3D. Most tools bolt a model onto one step of the flow; here every stage is generated rather than drawn. Edit by chat, export to KiCad or Gerber when it is ready to fabricate. Runs in the browser with nothing to install, and it is free. Built at RoboDIB.

    Electronic design automation · pcbeditor.com

§02Published resultApril 2026

Expert routing in Qwen3-Coder-Next is close to uniform

Profiling all 48 MoE layers with router-only probes gives a routing Gini of 0.1006 — the top 10% of experts take 13.22% of routing mass where uniform routing would give 10%. Pruning 512 experts to 128 retains 99.34% of observed routing. The full sweep took 4.98 seconds on CPU.

Gini
0.1006
Coverage
99.34%
Layers
48
Experts
512 → 128
Selections
491,520
Elapsed
4.98 s
§03Research
  • 2026Result

    Random-probe expert rankings degrade generation at 3% pruning

    Negative result, published here · 7 min

    Dropping just 2 of 64 experts per layer in OLMoE-1B-7B — a 3.1% reduction with top-k unchanged — produces measurable output degradation when the keep-list is chosen from random-probe statistics. One prompt collapses into a repetition loop while others show only wording drift.

  • 2026Result

    Pruning a Mixture-of-Experts checkpoint without a GPU

    Systems result, published here · 8 min

    Serving frameworks allocate every expert tensor on the GPU before any offload path engages, which makes large MoE checkpoints unprunable on modest hardware. Rewriting the checkpoint directly in safetensors, shard by shard, removes the constraint: OLMoE-1B-7B went from 13.0 GB to 7.39 GB in 7.6 seconds on CPU.

  • 2026Result

    Expert routing in Qwen3-Coder-Next is close to uniform

    Experimental result, published here · 6 min

    Router-only profiling of all 48 MoE layers of Qwen3-Coder-Next gives a routing Gini of 0.1006. Pruning 512 experts to 128 retains 99.34% of observed routing mass; the full sweep runs in 4.98 seconds on CPU.

  • 2024Paper

    Cognitive Analysis of Emotion from Voice Using Deep Learning

    Artificial Intelligence Research · 18 min

    Bridges cognitive models of how humans hear emotion with deep learning systems that classify it, and proposes an architecture in which the second is made accountable to the first.

§04Notes
  • What a voice says before the words do

    A speech system can transcribe perfectly and still handle a call badly, because the words were fine and the caller was furious. Notes on why affect belongs in the pipeline, and why a confidence score is not an explanation.

    6 min · Speech · Emotion recognition · Voice systems

  • Profiling a MoE model without loading it

    A 48-layer, 512-expert checkpoint profiled in 4.98 seconds on CPU. The trick is that expert routing is decided by a few megabytes of router weights, not by the hundreds of gigabytes behind them.

    5 min · Mixture of Experts · Inference · Tooling

  • Expert routing is flatter than you think

    Profiling all 48 layers of Qwen3-Coder-Next gives a routing Gini of 0.1006. Almost nothing is dead, almost nothing dominates — which changes what pruning a Mixture-of-Experts checkpoint can and cannot buy you.

    7 min · Mixture of Experts · Routing · Model compression

§05Code
§06Elsewhere

I also give talks on the JVM and distributed systems, write about engagement systems in healthcare, and keep a CV with the full history. I take on consulting work in ML systems and architecture.