---
name: Sadiq
tagline: "22 · A human with consciousness"
email: sadiq.build@gmail.com
socials:
  x: https://x.com/Md_Sadiq_Md/
  linkedin: https://www.linkedin.com/in/md-sadiq-md/
  github: https://github.com/MdSadiqMd/
skills:
  - name: "reinforcement learning"
  - name: "mech interp"
    tools: ["TransformerLens", "SAEs", "circuitsVis"]
  - name: "ai"
    tools: ["RAG", "agents", "memory systems", "evals"]
  - name: "zero knowledge"
    tools: ["zkVMs", "circom", "noir", "MPC"]
  - name: "high-performance web"
    tools: ["wasm", "webgpu", "offscreencanvas"]
  - name: "backend"
    tools: ["go", "rust", "python", "elixir", "typescript"]
  - name: "neural rendering"
    tools: ["neRF", "3D gaussian splatting"]
  - name: "solana"
  - name: "media processing"
    tools: ["ffmpeg", "image and audio processing"]
  - name: "infra"
    tools: ["MLOps", "terraform", "kubernetes", "nix", "AWS"]
---


# Sadiq
22 · A human with consciousness
## Summary
I work across a wide range of disciplines, including reinforcement learning, mech interp (TransformerLens, SAEs, circuitsVis), ai (RAG, agents, memory systems, evals), zero knowledge (zkVMs, circom, noir, MPC, ), high-performance web (wasm, webgpu, offscreencanvas), backend (go, rust, python, elixir, typescript), neural rendering (neRF, 3D gaussian splatting), solana, media processing (ffmpeg, image and audio processing) and infra (MLOps, terraform, kubernetes, nix, AWS)

I believe the hardest problems are often solved through interdisciplinary approaches (inspired by how openai solved the [planar unit distance problem](https://openai.com/index/model-disproves-discrete-geometry-conjecture)). A few explorations across disciplines:
## Skills

- **reinforcement learning**
- **mech interp**: TransformerLens, SAEs, circuitsVis
- **ai**: RAG, agents, memory systems, evals
- **zero knowledge**: zkVMs, circom, noir, MPC, 
- **high-performance web**: wasm, webgpu, offscreencanvas
- **backend**: go, rust, python, elixir, typescript
- **neural rendering**: neRF, 3D gaussian splatting
- **solana**
- **media processing**: ffmpeg, image and audio processing
- **infra**: MLOps, terraform, kubernetes, nix, AWS

## Projects

### Embodied Vision

Egocentric hand-tracking annotation platform with difficulty-aware frame classification.

- **Tags**: TypeScript, Python, Go, Computer Vision, Hand Tracking
- **Link**: https://github.com/MdSadiqMd/Embodied-Vision

### Relay

Version control for semantic search with temporal Merkle-committed epochs.

- **Tags**: Python, Semantic Search, Merkle Tree, Vector Database
- **Link**: https://github.com/MdSadiqMd/Relay

### CatalystSVM

Latency-aware ZK batch proving for SVM-based blockchains using SP1.

- **Tags**: Rust, TypeScript, ZK, SP1, Solana
- **Link**: https://github.com/MdSadiqMd/CatalystSVM

### AV Chaos Monkey

Chaos monkey for audio-video testing — simulates 1500+ WebRTC participants with network chaos injection.

- **Tags**: Go, WebRTC, UDP, Nix, Kubernetes, Chaos Engineering
- **Link**: https://github.com/MdSadiqMd/AV-Chaos-Monkey

### Rockbox

Sandbox environment built from bare bones using Nix.

- **Tags**: Rust, Elixir, Nix, Sandbox
- **Link**: https://github.com/MdSadiqMd/rockbox

### TraceZero

Privacy proxy for crypto transactions using ZK proofs.

- **Tags**: Rust, TypeScript, Circom, ZK, Solana, Privacy
- **Link**: https://github.com/MdSadiqMd/TraceZero

### neRF

Implementing Neural Radiance Fields papers from scratch.

- **Tags**: Python, Computer Vision, Neural Rendering, Deep Learning
- **Link**: https://github.com/MdSadiqMd/neRF

### TransformerLens

Open source contributions to mechanistic interpretability of GPT-style language models.

- **Tags**: Python, Mechanistic Interpretability, Transformers, Open Source
- **Link**: https://github.com/MdSadiqMd/TransformerLens

### Multi-Party Computation

Shamir's Secret Sharing scheme for Solana transaction signing.

- **Tags**: Rust, Python, MPC, Solana, Cryptography
- **Link**: https://github.com/MdSadiqMd/Multi-Party-Computation

### CacheFn

Plug-and-use, globally available Cargo cache package.

- **Tags**: Rust, Cache, Crates.io
- **Link**: https://github.com/MdSadiqMd/CacheFn

### Text-to-Shader

Generate GLSL shader code from natural language prompts using Gemini.

- **Tags**: Elixir, AI, Shaders, Gemini
- **Link**: https://github.com/MdSadiqMd/Text-to-Shader

### Snipzy Worker

Serverless video processing service using ffmpeg and yt-dlp.

- **Tags**: Go, Serverless, FFmpeg, Video Processing
- **Link**: https://github.com/MdSadiqMd/Snipzy-Worker

### Search Service

High-performance document search engine in Go — sub-second queries on 10M records.

- **Tags**: Go, Search, Full-Text Search
- **Link**: https://github.com/MdSadiqMd/search-service

## Experience

| Company | Role | Location | Period |
|---------|------|----------|--------|
| [Dognosis](https://dognosis.tech) | R&D Engineer | Bengaluru, India | 2025-10 – Present |
| [scale.jobs](https://scale.jobs) | Founding Engineer | Hyderabad, India | 2024-09 – 2025-09 |

### Dognosis — Details

- Solving cancer prediction by leveraging dogs olfactory capabilities, delivering accurate diagnostics from video-based sniff analysis
- Built multi-stage video processing pipeline YOLO-based dog tracking and ROI detection, centroid interpolation at 60 FPS, and stabilized 256x256 follow-cam crop generation via FFmpeg for FERAL(V-JEPA 2) model inference
- Trained and deployed V3 Fusion LSTM architecture combining kinematic time-series (head position, velocity, acceleration) with scalar context features (sniff duration, session progress, tortuosity)
- Managed Raspberry Pi camera fleet for automated daily session capture by debugging streaming protocols, service orchestration, and multi-network connectivity for reliable field deployment
- Managed cloud and on-prem infrastructure by automated data ingestion, processing, and backup workflows across distributed storage and compute environments
- Built multi-modal sensor synchronization matching high-frequency biosignals with video timelines, enabling time-aligned feature extraction for model training pipelines
- Overall worked with Python, OpenCV, YOLO, Airflow, TimescaleDB, PostgreSQL, Linux, FFmpeg, Docker, AWS, Rust

### scale.jobs — Details

- Only developer in the company besides the CTO
- Wrote AI content generation pipelines using LangChain and OpenAI SDK, while integrating Langfuse for prompt versioning, tracing, evaluation, and production monitoring
- Built AI Interviewer that conducted conversational mock interviews different for each and every job role, generated personalized feedback
- Optimized frontend performance using WebAssembly and TanStack, significantly improving rendering speed and overall user experience for large interactive applications
- Overall worked with TypeScript, tRPC, Hono, Prisma, PostgreSQL, QStash, LangChain, Langfuse, OpenAI SDK, TanStack, WebAssembly(WASM), and PostHog

## Contact

- **Email**: sadiq.build@gmail.com
- **x**: https://x.com/Md_Sadiq_Md/
- **linkedin**: https://www.linkedin.com/in/md-sadiq-md/
- **github**: https://github.com/MdSadiqMd/