# Dibyaprakash Pradhan

> Builds AI foundation architectures and models at AntEngage — designed from first principles rather than derived from the transformer lineage, for efficiency and capability the current generation cannot reach architecturally. Also builds pcbeditor.com at RoboDIB: describe a circuit in plain English, get a verified schematic and a fabricable board, free in the browser, exporting to KiCad and Gerber. Founder & CTO at AntEngage, Bengaluru, India. Published research on mixture-of-experts routing, checkpoint compression and speech emotion recognition.

This file is a machine-readable index of diby.app. The full text of every
page is available at https://diby.app/llms-full.txt.

## Identity
- Name: Dibyaprakash Pradhan (also written Dibya Prakash Pradhan; goes by Diby)
- Role: Founder & CTO at AntEngage (https://antengage.com/) — AI foundation architectures and models
- Also: Founder at RoboDIB (https://robodib.com), where pcbeditor.com is built
- Location: Bengaluru, India
- Email: dibyaprakash@robodib.com
- Profiles: https://www.linkedin.com/in/dibyaprakash-pradhan/, https://github.com/dibyapp, https://twitter.com/dibyapp, https://www.youtube.com/@dibyaprakash-pradhan, https://peerlist.io/dibyaprakash, https://www.f6s.com/member/dibyaprakash-pradhan, https://www.crunchbase.com/person/dibyaprakash-pradhan, https://www.producthunt.com/@dibyapp, https://instagram.com/dibyapp, https://www.quora.com/profile/Dibyaprakash-Pradhan

## Open source projects
- [MoE-Watcher-Modifier](https://diby.app/open-source/moe-watcher-modifier): MoE-Watcher-Modifier is an open-source, model-agnostic toolkit that analyses which experts a Mixture-of-Experts model actually uses, ranks them by importance, and rewrites the checkpoint with fewer experts — shrinking the model without touching training or inference code. Python, 5 stars, 2 forks. Source: https://github.com/dibyapp/MoE-Watcher-Modifier
- [Multitenancy in Spring Boot](https://diby.app/open-source/spring-boot-multitenancy): mt-base-service is an open-source Spring Boot reference implementation of multitenancy, covering tenant registration, per-tenant datasource resolution and Flyway-managed schema migration. Java, 1 stars, 7 forks. Source: https://github.com/dibyapp/mt-base-service
- [Location from IP Address](https://diby.app/open-source/location-from-ip-address): location-from-ip-address-service is an open-source Spring Boot service that resolves an IP address to a geographic location through a GeoIP lookup and renders the result on a map. Java, 0 stars, 0 forks. Source: https://github.com/dibyapp/location-from-ip-address-service
- [RSS GitHub Notifier for Slack](https://diby.app/open-source/rss-github-notifier-for-slack): rss-notifier-for-slack is a small open-source Java utility that watches GitHub activity over RSS and posts updates into Slack channels. Java, 0 stars, 0 forks. Source: https://github.com/dibyapp/rss-notifier-for-slack

## Research
- [Random-probe expert rankings degrade generation at 3% pruning](https://diby.app/research/random-probe-rankings-degrade-generation): 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. 2026, Negative result, published here.
- [Pruning a Mixture-of-Experts checkpoint without a GPU](https://diby.app/research/pruning-moe-checkpoints-without-a-gpu): 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. 2026, Systems result, published here.
- [Expert routing in Qwen3-Coder-Next is close to uniform](https://diby.app/research/expert-routing-qwen3): 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. 2026, Experimental result, published here.
- [Cognitive Analysis of Emotion from Voice Using Deep Learning](https://diby.app/research/cognitive-analysis-emotion-voice): 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. 2024, Artificial Intelligence Research.

## Notes
- [What a voice says before the words do](https://diby.app/notes/what-voice-says-before-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. (2026-06-02, 6 min)
- [Profiling a MoE model without loading it](https://diby.app/notes/profiling-moe-models-without-loading-them): 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. (2026-05-14, 5 min)
- [Expert routing is flatter than you think](https://diby.app/notes/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. (2026-04-21, 7 min)

## Writing
- [Smart Triage: The Cure for Spam Calls and Lost Patients](https://blog.antengage.com/smart-triage-for-clinics): Spam, wrong leads and wasted time — why your front desk needs something better. (2025-10-28)
- [The Patient Engagement Maze: Quality vs Quantity](https://blog.antengage.com/the-patient-engagement-maze-quality-vs-quantity): Every clinic tries the same fixes. Each one opens another leak. Play the maze — each move looks like progress, and each one has a cost. (2025-09-29)
- [The Follow-Up Leak After the Visit — "India clinic reality check"](https://blog.antengage.com/missed-follow-ups-clinic-india): Addressing follow-up challenges in Indian clinics: identifying the leaks and implementing solutions for better patient outcomes. (2025-08-18)

## Products
- [AntEngage](https://antengage.com): AI foundation architectures and models, built from scratch. Role: Founder & CTO. Site: https://antengage.com
- [PCBEditor](https://diby.app/products/pcbeditor): PCBEditor is a free, browser-based AI circuit design tool that turns a plain-English description into a complete printed circuit board — generating the schematic, component footprints, netlist and routed traces, then running AI design rule checking and copper pour, with 3D layer stack-up and board visualisation, and exporting to KiCad or Gerber. Role: Founder at RoboDIB. Site: https://pcbeditor.com
- [RoboDIB](https://robodib.com): Robotics and hardware, and the home of PCBEditor. Role: Founder. Site: https://robodib.com
- [Fonix AI](https://fonix.ai): Voice and messaging infrastructure for AI agents. Role: Builder. Site: https://fonix.ai
- [AIVoice Live](https://aivoice.live): Real-time voice AI. Role: Builder. Site: https://aivoice.live

## Talks and shorts
- [What is gRPC — HTTP vs gRPC for your next API](https://diby.app/videos/grpc-vs-http-for-your-next-api): HTTP is universal, human-readable and understood by every tool in the stack. gRPC is faster on the wire, strongly typed at the boundary, and awkward everywhere a browser is involved. This conversation works through which properties actually decide the choice for a new API. (talk, 2024-04-04, GeekyAnts)
- [Diving deep into GraalVM — native images with Spring Boot](https://diby.app/videos/graalvm-native-images-spring-boot): GraalVM compiles a JVM application ahead of time into a native executable. Startup drops from seconds to milliseconds and memory footprint falls sharply, which changes what the JVM is viable for. The cost is a closed-world assumption that breaks the reflective tricks much of the Java ecosystem is built on. (talk, 2023-09-19, GeekyAnts)
- [Resolving a location from an IP address in Java](https://diby.app/videos/ip-address-geolocation-in-java): Turning an IP address into a country, region and city is a routine requirement for analytics, fraud checks and localisation. Doing it against a local GeoIP database rather than a hosted API keeps the lookup fast, free per request, and keeps user IP addresses inside your own infrastructure. (talk, 2020-05-27, Programmatic DIB)
- [Camera text recognition on Android — building a Google Lens-style OCR app](https://diby.app/videos/android-camera-text-recognition-ocr): Reading text off a live camera feed is a pipeline problem rather than a single API call: acquire frames, hand them to a recognition engine, and reconcile a stream of noisy per-frame results into something stable enough to show a user. (talk, 2020-06-05, Programmatic DIB)
- [Which language should you use for Android development?](https://diby.app/videos/best-language-for-android-development): Android can be written in Java, Kotlin, C++ through the NDK, or any of several cross-platform frameworks. The languages differ less in capability than in how much ceremony they demand and how well they match the platform's own direction. (talk, 2020-05-25, Programmatic DIB)
- [Python CGI programming — the architecture, and a first program](https://diby.app/videos/python-cgi-programming-architecture): CGI is the oldest way to make a web server run a program, and understanding it explains why every framework since has been built the way it is. One process per request is a beautifully simple model with exactly one fatal property. (talk, 2020-05-31, Programmatic DIB)
- [Why vLLM?](https://diby.app/videos/why-vllm): Serving a language model well is mostly a memory management problem. vLLM's contribution is treating the KV cache like virtual memory — paging it — which raises how many requests you can hold concurrently far more than a faster kernel would. (short, 2026-03-15, Programmatic DIB)
- [This is why quantization matters](https://diby.app/videos/why-quantization-matters): Quantization stores weights at lower precision — 8-bit or 4-bit instead of 16. The obvious win is that the model fits. The less obvious and often larger win is that inference is memory-bandwidth bound, so moving fewer bytes per token makes it faster too. (short, 2026-03-18, Programmatic DIB)
- [Training an AI model is slow — back-propagation explained](https://diby.app/videos/why-training-is-slow-backpropagation): Training is slow because every step runs the network forwards, then runs a second pass backwards to compute how every parameter should change, and the backward pass needs the forward pass's intermediate values kept in memory the whole time. (short, 2026-03-20, Programmatic DIB)
- [Gradient descent explained](https://diby.app/videos/gradient-descent-explained): Gradient descent is the whole of learning reduced to one instruction: measure which direction increases the error, and take a small step the other way. Everything else in an optimiser is a refinement of how large that step should be. (short, 2026-03-23, Programmatic DIB)

## Consulting services
- General Consulting: Advisory work on technology decisions, delivery and direction.
- Spring Boot Training: Professional training and mentoring for Java and Spring Boot teams.
- Software Architecture: System design and architecture guidance for products under load.
- Software Strategy: Strategic technology planning and roadmaps.
- Team Building Strategy: Building and scaling high-performing engineering teams.
- Architecture Review: Expert review and optimisation of existing systems.
- Security Review: Comprehensive security assessment and guidance.

## Pages
- [Index](https://diby.app/): latest result and selected work
- [Research](https://diby.app/research): papers and experimental results, with figures and reproduction commands
- [Notes](https://diby.app/notes): technical notes on routing, compression and speech
- [Code](https://diby.app/open-source): open source projects, one documentation page each
- [Talks](https://diby.app/videos): talks and shorts, each with its own page of written notes
- [About](https://diby.app/about): background and current focus
- [CV](https://diby.app/experience): full role-by-role work history
- [Writing](https://diby.app/writings): essays on engagement systems in healthcare
- [Products](https://diby.app/products): startups and platforms
- [Consulting](https://diby.app/hire-me): how I work with teams
- [Contact](https://diby.app/contact): how to get in touch
- [Gargantua](https://diby.app/gargantua): an interactive Schwarzschild black hole rendered by integrating photon geodesics, with a write-up of the physics and implementation

## Optional
- [RSS feed](https://diby.app/feed.xml)
- [Sitemap](https://diby.app/sitemap.xml)