Skip to content
View ManoharPaturi's full-sized avatar

Highlights

  • Pro

Block or report ManoharPaturi

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ManoharPaturi/README.md
Manohar Paturi Banner
Typing SVG



Coordinates   Open Source   Projects   Tech Arsenal   Telemetry   Connect



⚡ Coordinates & Engineering Philosophy

🛰️ manohar.yaml

identity: Manohar Paturi
institution: Amrita Vishwa Vidyapeetham
degree: B.Tech in Artificial Intelligence Engineering
core_domains:
  - Reinforcement Learning (Dueling Double DQN, PER, NoisyNets)
  - Biomedical Signals (EEG artifact subspace isolation via SVD)
  - 3D Computer Vision & Biomechanics (Stereo DLT, Three.js)
  - Offline-First Mobile CV (OpenCV) & POSIX Systems (C)
hardware_rig:
  - 2-Laptop Stereo DLT Motion Capture Rig (Real-Time 3D)
  - Emotiv EPOC X 14-Channel Wireless EEG Headset
status: "⚡ Training agents · Denoising biosignals · Shipping PRs"

🧭 First Principles

Build it from scratch at least once
The magic disappears, the understanding stays.

If you can't reproduce it, you don't understand it yet.
Read the paper, derive the math, implement from blank buffers.

Real data over clean examples
Messy inputs and hardware artifacts teach what tutorials hide.

Know what's underneath your abstractions
Every high-level framework leaks eventually.

Author specs in millimetres
Let code, grids, and schemas compile from ground truth.


🌐 Open Source Impact — 29 Upstream PRs across 15 Global Orgs

"I've shipped fixes into repos with more stars than I have followers — working on flipping that ratio too."
Upstream fixes tackling concurrency race conditions, WCAG accessibility compliance, inference pipeline DX, float16 gradient underflows, and security tooling.

Meta Google DeepMind Microsoft Anthropic AWS Hugging Face vLLM Ollama Keras Mistral AI LangChain Weaviate deepset Haystack EleutherAI ComfyUI


🏆 Selected Upstream Highlights

Organization Target Repository Fix Shipped & Architectural Significance PR Status
Meta facebook/astryx Added elevation prop for ToggleButton — survived full review loop with Meta's design-system team #6037 Merged
LangChain langchain-ai/langchain Pruned stale Args / Raises docstrings from core callbacks & generation pipeline #40211 Merged
Meta facebook/astryx Guard TextInput.onEnter against premature Japanese-IME conversion commits #6083 Open
Meta facebook/astryx Enforced 44px WCAG touch-target floor on SegmentedControlItem for coarse pointers #6084 Open
deepset deepset-ai/haystack Cross-batch write–write concurrency race condition in agent tool scheduler resolved by LLM call order #12628 Open
Keras keras-team/keras Fixed -inf gradients in normalize() under float16 by computing L2 norm in float32 #23566 Open
vLLM vllm-project/vllm Packaged numactl into ROCm container images preventing --numa-bind silent no-ops #55433 Open
Ollama ollama/ollama Normalized escaped pattern literals in structured format schemas at llama-server boundary #18248 Open
Microsoft microsoft/autogen Prevented stale hunk corruption in TextCanvas.apply_patch via strict context validation #8195 Open
Microsoft microsoft/PyRIT Formulated Garak LLM red-teaming scenario (Jinja2 template injection + SQLi echo) #2576 Open
Anthropic anthropics/claude-code-action Skip malformed buffered comment lines cleanly instead of failing the CI post step #1797 Open
ComfyUI Comfy-Org/ComfyUI Mask-editor painted uploads no longer destroy the original image under the mask #16141 Open
📂 View Full Upstream Contribution Ledger (17 Additional PRs)
Target Repository Subsystem / Contribution PR Status
huggingface/transformers Bypass symlinked hub-cache test on platforms lacking symlink permissions #48530 Open
weaviate/weaviate Correct object_count Prometheus metric description help text #12951 Open
weaviate/weaviate Add missing rq-4 to ALLOWED_COMPRESSION_TYPES schema specification #12950 Open
google-deepmind/mujoco Correct inconsistent axis labelling in solver impedance (solimp) physics figures #3548 Open
google-deepmind/mujoco Physics engine documentation accuracy sweep #3550 Open
microsoft/autogen Multi-agent framework documentation accuracy across 16 files #8196 Open
aws/aws-cdk Drop stale sep reference from ArnComponents.arnFormat default documentation #38775 Open
ollama/ollama Log provenance trace of dynamically loaded model context length #18249 Open
Comfy-Org/ComfyUI Fix Trellis2UpsampleStage hardcoded lr_resolution for 64-grid spatial decodes #16114 Open
EleutherAI/lm-evaluation-harness Verified evaluation benchmark specifications across 16 task suites #4102 Open
facebook/astryx Render inline anchor links correctly so parent Text containers apply truncation #6038 Open
anthropics/claude-code-action Extract user trigger prompt strictly on word-boundary regex patterns #1799 Open
huggingface/datasets Fix documentation guide template configuration #8567 Open
mistralai/mistral-common Fix experimental guide parser configuration #306 Open

🚀 Deep Tech Lab — One Project per Problem Domain

Every project here was chosen because the interesting challenge lives in the underlying mathematical and physical subject matter, not in boilerplate frameworks.

domain project what's under the hood
🧠 reinforcement learning Tic-Tac-Toe × Double DQN Double Q-learning targets (overestimation bias fix) + dueling value/advantage streams + NoisyNet exploration + prioritized replay — every trick implemented from its paper, trained purely by self-play
dynamical systems DMD power oscillations Regularized Dynamic Mode Decomposition on 29-generator PMU recordings — recovers oscillation-mode frequency, damping ratio and generator participation factors
🩺 biomedical signals SVD EEG denoising EOG artifacts isolated as a low-rank subspace of real Emotiv EPOC X recordings via SVD — reproducing Sadasivan & Dutt (1996)
🎥 3D vision · biomechanics mocapX1 Markerless motion capture end-to-end — pose → stereo DLT triangulation → filtering → joint kinematics → JSON/CSV/TRC/C3D/BVH exports, with a FastAPI + Three.js web UI
🧬 graph theory · bioinformatics T1D protein network Degree / betweenness / closeness centrality over the Type-1-Diabetes protein interactome from STRING data — ranks candidate hub genes with NetworkX
🔌 computer networks OSI model simulator Five networked processes (clients, switch, router, server) passing real socket traffic through per-layer encapsulation/decapsulation
🖥️ systems programming File manager on raw syscalls Dual-pane GTK3 file manager where listing, copy, recursive delete and zip run directly on POSIX system calls + libzip
📱 mobile · Android projX — OMR evaluation Android-first Flutter app that grades OMR answer sheets fully offline — camera → on-device OpenCV bubble detection → versioned answer keys → PDF/XLSX marksheets, all derived from one sheet spec authored in mm
🌐 full-stack · client–server SRMS seating Seat & venue reservation built twice over one shared domain model — a Java Swing desktop client and a Spring REST API, both against MySQL persistence
more on the shelf
  • VS7.1 mocap — dual-laptop real-time multi-person capture with stereo 3D reconstruction, kinematics engine & validation pipeline
  • MOCAP_MANU — EasyMocap + SMPL monocular body fitting with an upload-to-analyze web app
  • C Data Structures — header-only C++17 heaps, BSTs, graphs with BFS/DFS, each with an interactive demo
  • Chef Tony — Gemini 2.0 Flash recipe chatbot that finds its own dish photos

🛠️ Tech Arsenal

👨‍💻 Programming Languages

Languages

🧠 AI, Computer Vision & Signal Processing

AI & CV

MediaPipe Hugging Face NetworkX NumPy Pandas SciPy SMPL

🌐 Mobile & Full-Stack Engineering

Frameworks

GTK3 ZeroMQ

⚙️ Systems, Cloud & Data Infrastructure

Infra & Data

📈 GitHub Telemetry & Live Analytics


📬 Direct Uplink

Interested in collaborating on complex AI, biosignals, or systems problems?


Email   GitHub   LinkedIn



Footer Wave

⚡ Built from first principles · $ exit 0

Pinned Loading

  1. Chatbot-for-recepies-with-images Chatbot-for-recepies-with-images Public

    Chef Tony — desktop recipe chatbot (Python + tkinter) powered by Gemini 2.0 Flash, with automatic dish photos from DuckDuckGo image search

    Python

  2. D9_MFC4_SVD-BASED-NOISE-REDUCTION-IN-EEG-SIGNALS- D9_MFC4_SVD-BASED-NOISE-REDUCTION-IN-EEG-SIGNALS- Public

    SVD-based removal of EOG artifacts from real EEG recordings (Emotiv EPOC X) — a linear-algebra approach to biomedical signal enhancement, in MATLAB

  3. File-Management-System-Using-Basic-System-Calls-in-C File-Management-System-Using-Basic-System-Calls-in-C Public

    Dual-pane GTK3 file manager in C — listing, mkdir, rename, recursive delete, copy and zip implemented directly on POSIX system calls + libzip

    C 1

  4. mocap_mac mocap_mac Public

    VS7.1 real-time multi-person motion capture with dual-laptop stereo 3D reconstruction — DLT triangulation, kinematics engine, validation pipeline, and Plotly reports

    HTML

  5. MOCAP_MANU MOCAP_MANU Public

    Monocular markerless motion capture workspace — vendored EasyMocap + SMPL fitting, preset CLI, and a FastAPI upload-to-analyze web app

    Python

  6. TIC-TAC-TOE-Player-By-DOUBLE-DQN-MODEL TIC-TAC-TOE-Player-By-DOUBLE-DQN-MODEL Public

    Tic-Tac-Toe RL — Double Dueling DQN with NoisyNet exploration and Prioritized Experience Replay trained by self-play, plus an unbeatable minimax (tkinter & raylib)

    Python