itachi./ 1824

MOHAMMED AFFAN / SOFTWARE DEVELOPER

itachi.

Software developer.
AI systems & infrastructure.

I build agents, model integrations, and the infrastructure they run on — across cloud services, local AI, and applications.

Explore my work View resume ↗
Currently building at
Explore the work

Scoped tools. Streamed responses.

Polli connects caller context, repository retrieval, and tools across its agent interfaces.

Explore Pollinations work

Quick overview ↗ Experience ↗Try the scene editor ↗

01 / POLLINATIONS / AGENT SYSTEMS

Meet Polli.

I build across Polli’s agent interfaces, tools, and execution.
Explore a request, its caller boundary, and its result.

Read the case study ↗
HTTP API

Where is request authentication handled?

EXECUTION PATHREADY
TRACE

Run the request, or select a step to inspect it.

RESULT PREVIEW

api/server.py reads the authorization header and establishes caller context. core/request_auth.py carries request authorization; ai/tool_filters.py controls available tool actions.

SOURCE MAPapi/server.pyReads the authorization header; sets HTTP caller context.core/request_auth.pyCarries request authorization in a ContextVar.ai/tool_filters.pyFilters tool actions for the caller.

Local walkthrough · Prepared examples · No live model or GitHub action.

DISCORD / GITHUB / OPENAI-COMPATIBLE APIS

Selected work / 02 / ML Research

Ouroboros

Ask better
questions.

I develop language-model experiments in PyTorch and compare proposed mechanisms against explicit baselines.

PythonPyTorchML researchEvaluation
Case study & evidence ↗

Research · In development

INTERACTIVE EXHIBIT
OUROBOROS / INDEPENDENT RESEARCHIn development

The question

Does a proposed change survive a controlled comparison?

I develop language-model prototypes in PyTorch, compare them with explicit baselines, and document where the evidence is limited.

Work in progress
Local experiments, baseline comparisons, and review of limitations and negative findings.
Public material
A research log and a high-level account of the experimental process.
Publication
Papers are planned. No publication or measured performance claim is presented here.
Read the public research log

Unpublished methods, architecture, and training artifacts remain private.

PUBLIC RESEARCH OVERVIEWEngineering decisions

Selected work / 03 / Local AI / Distributed Systems

Sylph

Intelligence,
closer to home.

I am building a Rust runtime for local inference and exploring peer-to-peer exchange of model adapters.

RustLocal inferenceLoRAPeer-to-peer
Case study & evidence ↗

Research · In development

INTERACTIVE EXHIBIT
LOCAL RUNTIMEsample-base-A
INCOMING OFFERsample-base-A
Change the base model or edit the payload, then check what the receiver can accept.

Real SHA-256 over synthetic text, computed in your browser. This explains two independent checks; it does not run Sylph, load an adapter, or establish model safety.

EXPLORE THE ARCHITECTUREEngineering decisions

Selected work / 04 / Agent Infrastructure

Memory-Arc

A conversation
with a memory.

I built a Python memory engine with configurable processing modes, persistent retrieval, and asynchronous interfaces.

PythonQdrantAsyncIOVector retrieval
Case study & evidence ↗

Public source

INTERACTIVE EXHIBIT
Memory-Arc / Context experimentLocal example
THE FIRST CONVERSATION

Deploy the API to Cloudflare Workers. Keep the database in D1.

A useful detail is sitting in the recent conversation.
PERSISTENT MEMORY

Nothing saved yet.

Prepared conversations illustrate persistence and recall. This browser example does not run the Python engine or vector search.

EXPLORE THE ARCHITECTUREEngineering decisions

Selected work / 05 / ML Tooling

Agent evaluation environments

Put the agent
to the test.

I built OpenEnv investigation environments with procedural cases, MCP tools, evolving state, and adversarial tests.

PythonOpenEnvMCPDocker
Case study & evidence ↗

Hackathon projects

INTERACTIVE EXHIBIT
Investigation desk / Sample caseIncident open
CASE / OUTBOUND CONNECTION

A workstation contacts an unfamiliar host.

Inspect the evidence. Decide whether the incident can be closed.

The alert alone does not establish what happened.

A simplified local exercise. Its rule is illustrative; it does not reproduce the OpenEnv reward or execute security tools.

EXPLORE THE ARCHITECTUREEngineering decisions

Selected work / 06 / Systems / Developer Tools

Repository intelligence

Know the code.
Then change it.

I am developing local Rust tools for repository discovery, structured retrieval, and validation context.

RustSQLite / FTS5Tree-sitterCLI
Case study & evidence ↗

Private · In development

INTERACTIVE EXHIBIT
Source explorer / Synthetic repositorySource current
src/auth.tsINDEXED
export function authorize(caller) {
  return caller.role === "maintainer";
}

Search finds an answer. Changing the source makes its validation matter.

Local string search over synthetic files. The private Rust implementation uses Tree-sitter and SQLite/FTS5.

EXPLORE THE ARCHITECTUREEngineering decisions

Selected work / 07 / Full-Stack / Mobile / 3D

SceneForge

Edit the scene.
See the change.

I built a scene editor that turns streamed JSON into editable 3D geometry.

Next.jsFlutterSupabaseThree.js
Case study ↗

Private projects

INTERACTIVE EXHIBIT
SceneForge / Scene studyInteractive 3D
Drag to orbit · Click to inspect
{ scene: "pavilion", format: "typed objects" }
Preparing sceneLocal sample scenes · No model calls
SceneForge / Local scene editor

Change the scene.

Edit object types, positions, scales, or colors. Apply the description to rebuild the scene.

Local sample data. This preview demonstrates structured scene rendering, without an AI model.

EXPLORE THE ARCHITECTUREEngineering decisions

Experience / Building with teams

Inside
the work.

AI systems, infrastructure, and the workflows that connect them.

Resume ↗Detailed CV ↗
Feb 2025 – Present

Pollinations.ai ↗

Developer · MLOps & AI Infrastructure

I develop agent interfaces and multimodal MCP tools, maintain incremental repository retrieval, and connect streamed APIs to isolated execution.

PythonTypeScriptMCPCloudflareGitHub ActionsDocker
Inspect my contributions ↗
Feb 2025 – Present

Myceli.ai ↗

Developer · AI Systems & Automation

I develop automations and workflows connecting models, services, and agent modules across cloud and local environments.

AI infrastructureAutomationAgent systems

Behind the handle

Itachi online.
Affan in person.

I like understanding a system far enough down that I can change it. That takes me between model research, local AI, cloud infrastructure, and the interfaces people use.

Diploma · AI/ML pathway
MS Ramaiah PolytechnicPursuing · 5th semester

01

AI & agent systems

Custom MCP servers · Tool orchestration · RAG & vector search · Model fine-tuning · Local inference · PyTorch

See it in the work ↗
02

Infrastructure & delivery

Cloudflare Workers · Kubernetes · Docker · GitHub Actions · Linux · Model deployment

See it in the work ↗
03

Languages & applications

Python / FastAPI · TypeScript / JavaScript · Rust / SQL · React / Next.js · Dart / Flutter · Three.js

See it in the work ↗
04

Data & persistence

Cloudflare Vectorize · D1 / SQLite · Qdrant · Supabase · Embeddings · Incremental indexing

See it in the work ↗

Open to opportunities

What are
you building?

Software. AI. Infrastructure. Applications.
Tell me about the problem your team is working on.

1824.itachi@gmail.com ↗
GitHub ↗LinkedIn ↗Resume ↗CV ↗
Mohammed Affan · Bengaluru, India · 2026
Polli / Interactive walkthrough
Run the workflow

Follow the caller boundary.

Follow a request through caller context, tool visibility, retrieval, and response streaming. Click a step, connection, or trace to inspect it.

Execution trace / Click to inspect

Start the walkthrough, or inspect any step before running it.

Simplified from the project architecture. Runs locally; no live models, credentials, or external services.

Polli / Inspector

Workflow overview

Receive

An OpenAI-compatible request enters the API and receives a caller context.

Why it matters

Polli connects caller context, selected models, repository evidence, and tools. Exposing a tool is distinct from validating every action it can perform.

Synthetic payloads explain the boundaries. They are illustrative examples, not project API schemas or live production traces.

Ouroboros / Interactive walkthrough
Run the workflow

Follow a research decision.

A high-level research workflow. Model internals, implementation details, and unpublished results are intentionally omitted. Click a step, connection, or trace to inspect it.

Execution trace / Click to inspect

Start the walkthrough, or inspect any step before running it.

Simplified from the project architecture. Runs locally; no live models, credentials, or external services.

Ouroboros / Inspector

Workflow overview

Question

Define the question and what evidence would support it.

Why it matters

Only the public research process is shown. Unpublished architecture, experimental methods, training details, and measurements remain private.

Synthetic payloads explain the boundaries. They are illustrative examples, not project API schemas or live production traces.

Sylph / Interactive walkthrough
Run the workflow

Inspect an arriving adapter.

Inspect the compatibility and integrity gates around a received model-adapter update. Click a step, connection, or trace to inspect it.

Execution trace / Click to inspect

Start the walkthrough, or inspect any step before running it.

Simplified from the project architecture. Runs locally; no live models, credentials, or external services.

Sylph / Inspector

Workflow overview

Offer

A peer offers metadata describing a model-adapter delta.

Why it matters

Compatibility and integrity are separate gates. Passing them does not establish that an adapter is beneficial or safe in every composition; evaluation continues beyond these checks.

Synthetic payloads explain the boundaries. They are illustrative examples, not project API schemas or live production traces.

Memory-Arc / Interactive walkthrough
Run the workflow

Follow the context that persists.

Explore the difference between storing useful context and recalling it for a conversation. Click a step, connection, or trace to inspect it.

Execution trace / Click to inspect

Start the walkthrough, or inspect any step before running it.

Simplified from the project architecture. Runs locally; no live models, credentials, or external services.

Memory-Arc / Inspector

Workflow overview

History

The memory manager receives recent conversation context.

Why it matters

Recent thread history and persistent vector memory serve different purposes. Processing mode and provider configuration determine the application’s actual memory behavior.

Synthetic payloads explain the boundaries. They are illustrative examples, not project API schemas or live production traces.

Agent evaluation environments / Interactive walkthrough
Run the workflow

Follow the investigation.

Walk through an investigation and see why evidence matters to evaluation. Click a step, connection, or trace to inspect it.

Execution trace / Click to inspect

Start the walkthrough, or inspect any step before running it.

Simplified from the project architecture. Runs locally; no live models, credentials, or external services.

Agent evaluation environments / Inspector

Workflow overview

Scenario

A procedural scenario establishes the initial evidence and environment state.

Why it matters

These environments evaluate actions in a simulated scenario. Evidence gathering and the evolving state matter alongside the final answer.

Synthetic payloads explain the boundaries. They are illustrative examples, not project API schemas or live production traces.

Repository intelligence / Interactive walkthrough
Run the workflow

Check the evidence against the source.

Follow a code question from repository structure to evidence with a known source state. Click a step, connection, or trace to inspect it.

Execution trace / Click to inspect

Start the walkthrough, or inspect any step before running it.

Simplified from the project architecture. Runs locally; no live models, credentials, or external services.

Repository intelligence / Inspector

Workflow overview

Discover

The local repository supplies the source content for analysis.

Why it matters

A retrieved result and a validated claim about current source are different things. Source state is part of interpreting repository evidence.

Synthetic payloads explain the boundaries. They are illustrative examples, not project API schemas or live production traces.

Applications, end to end / Interactive walkthrough
Run the workflow

Follow the data into the interface.

Explore the application paths behind streamed scenes, editable presentations, and mobile reporting. Click a step, connection, or trace to inspect it.

Execution trace / Click to inspect

Start the walkthrough, or inspect any step before running it.

Simplified from the project architecture. Runs locally; no live models, credentials, or external services.

Applications, end to end / Inspector

Workflow overview

Request

The user describes a scene through the application interface.

Why it matters

These examples demonstrate distinct application paths: typed scene rendering, streamed presentation content, and records-to-document reporting.

Synthetic payloads explain the boundaries. They are illustrative examples, not project API schemas or live production traces.

ITACHI / PROJECT INDEX

Find your way.

Merged contributions

Polli

A connected agent platform spanning Discord, GitHub, and OpenAI-compatible APIs. Repository intelligence, multimodal tools, context management, and streamed execution inside Pollinations.

PythonMCPCloudflareGitHub Actions
Explore the project Engineering notes ↗
↑ ↓ to explore · Enter to open

Interactive exhibit