inference · cold start
inference v1.0 · status: open to work

01 · SYSTEM PROMPT

TODO(content): Nur Mohammad Sowan

TODO(content): AI-focused Software Engineer

TODO(content): I design and build AI systems end to end — retrieval, agents, evaluation, and the interfaces that make them usable. This portfolio is one of them: it boots like a model-serving stack, then renders itself into the work.

> The first instruction sets the behavior, tone, and expectations for everything that follows.

[view trace]
{
  "role": "system",
  "channel": "final",
  "tokens": 128,
  "temperature": 0.7
}

02 · FOUNDATION MODEL

About Me

> Core identity and base capabilities form the foundation every project builds on.

[view trace]
{
  "model": "identity-base",
  "context_window": 200000,
  "quantization": "none"
}

TODO(content): I am a software engineer focused on applied AI. I like the unglamorous parts — data plumbing, evals, latency budgets — because that is where reliable products are actually won.

TODO(content): My base capabilities span full-stack product engineering and modern ML tooling. I care about clear interfaces, honest metrics, and shipping something a real person can use on the first try.

location TODO(content): Dhaka, Bangladesh

03 · SKILLS & FRAMEWORKS

Skills / Tech Stack

> Frameworks and tools are the reusable building blocks used to construct reliable systems.

[view trace]
{
  "adapters": [
    "react",
    "astro",
    "python",
    "rag"
  ],
  "loaded": true,
  "warm": true
}
Languages
  • TODO(content): TypeScript
  • TODO(content): Python
  • TODO(content): SQL
AI / ML
  • TODO(content): RAG pipelines
  • TODO(content): Prompt orchestration
  • TODO(content): Model evaluation
  • TODO(content): Fine-tuning
Frameworks
  • TODO(content): React
  • TODO(content): Astro
  • TODO(content): Node.js
  • TODO(content): FastAPI
Infra & Tooling
  • TODO(content): Docker
  • TODO(content): Postgres + pgvector
  • TODO(content): Cloudflare
  • TODO(content): GitHub Actions

04 · PRE-TRAINING CORPUS

Education

> Foundational knowledge comes from the corpus that shaped the earliest behavior.

[view trace]
{
  "corpus": "cs-fundamentals",
  "shards": 12,
  "epochs": 4,
  "status": "indexed"
}

  1. TODO(content): University Name

    TODO(content): 2019 — 2023

    TODO(content): B.Sc. in Computer Science

    • TODO(content): Concentration in machine learning and distributed systems.
    • TODO(content): Thesis on retrieval-augmented question answering over course materials.

05 · FINE-TUNING DATA

Work Experience

> Real-world feedback refines broad capability into sharper, more useful behavior.

[view trace]
{
  "method": "sft",
  "examples": 2,
  "objective": "measurable-impact",
  "loss": 0.14
}

  1. TODO(content): Software Engineer · TODO(content): Company Name

    TODO(content): 2023 — Present
    • TODO(content): Built and shipped a retrieval pipeline serving production traffic with sub-200ms p95 latency.
    • TODO(content): Designed an evaluation harness that cut regression review time from hours to minutes.
    • TODO(content): Partnered with product to turn a research prototype into a launched feature.

  2. TODO(content): Software Engineering Intern · TODO(content): Company Name

    TODO(content): Summer 2022
    • TODO(content): Prototyped an internal tool later adopted by two teams.
    • TODO(content): Improved test coverage on a core service from 40% to 85%.

06 · KNOWLEDGE BASE

Featured Projects

> A knowledge base stores practical evidence that can be retrieved and applied.

[view trace]
{
  "query": "best work",
  "top_k": 3,
  "retrieved": [
    "navigate-me",
    "wayfinder",
    "coursework"
  ]
}

  • TODO(content): Navigate Me

    TODO(content): Indoor wayfinding that turns a building floor plan into turn-by-turn directions.

    • TODO(content): React
    • TODO(content): Python
    • TODO(content): Graph search
    • TODO(content): Cut average time-to-destination for first-time visitors in user testing.
    • TODO(content): Runs fully offline on a phone after first load.

    repo demo

  • TODO(content): Wayfinder

    TODO(content): A semantic search layer over personal notes with cited answers.

    • TODO(content): TypeScript
    • TODO(content): pgvector
    • TODO(content): OpenAI-compatible API
    • TODO(content): Answers include source citations to reduce hallucination risk.
    • TODO(content): Indexes 10k+ documents with incremental updates.

    repo

  • TODO(content): Coursework Systems

    TODO(content): A collection of academic systems projects — schedulers, parsers, and simulators.

    • TODO(content): C++
    • TODO(content): Python
    • TODO(content): Top-grade submissions across systems and algorithms coursework.
    • TODO(content): Documented and open-sourced for other students.

    repo

07 · VECTOR DATABASE

GitHub & Open Source

> A searchable index of code makes prior work easier to retrieve, inspect, and reuse.

[view trace]
{
  "index": "hnsw",
  "metric": "cosine",
  "dim": 768,
  "vectors": 2
}

TODO(content): A searchable trail of what I build in the open — pinned repositories, experiments, and contributions.

profile TODO(content): your-handle

08 · TOOLS

Certifications / Integrations

> External tools expand what the system can do beyond its base model.

[view trace]
{
  "tools_registered": 2,
  "schema": "openapi-3.1",
  "auth": "none"
}

  • TODO(content): Machine Learning Specialization

    TODO(content): Issuer / Platform

    TODO(content): 2023 view

  • TODO(content): Cloud Practitioner

    TODO(content): Cloud Provider

    TODO(content): 2022

09 · AGENT

Problem-Solving Philosophy

> An agent combines context, tools, and goals into deliberate action.

[view trace]
{
  "loop": [
    "observe",
    "plan",
    "act"
  ],
  "max_steps": 3,
  "guardrails": true
}
  1. 01

    TODO(content): Start from the failure mode

    TODO(content): I design backwards from how a system is most likely to break, then build the guardrails first.

  2. 02

    TODO(content): Measure before you tune

    TODO(content): No change ships without an eval that proves it helped. Vibes are a hypothesis, not a result.

  3. 03

    TODO(content): Ship the smallest honest version

    TODO(content): A narrow feature that works beats a broad one that almost does. Scope is a feature.

10 · EVALUATION

Achievements / Testimonials

> Evaluation provides external signals that the system performs well in practice.

[view trace]
{
  "suite": "peer+outcomes",
  "cases": 3,
  "passed": true
}

  • TODO(content): Hackathon Winner

    TODO(content): First place out of 40+ teams for an accessibility-focused navigation tool.

  • TODO(content): Open-source impact

    TODO(content): A utility library used by several downstream projects.

TODO(content): Ships fast without cutting the corners that matter — the kind of engineer you want owning the hard part.
TODO(content): Jane Doe TODO(content): Engineering Manager, Company Name

11 · API ENDPOINT

Contact

> An API endpoint is the public interface where other systems can connect.

[view trace]
{
  "method": "POST",
  "path": "/contact",
  "transport": "mailto",
  "status": 200
}

TODO(content): The fastest way to reach me is email. I reply within a day or two.

status TODO(content): Open to full-time and contract work in applied AI.

TODO(content): you@example.com LinkedIn GitHub