Skip to content

Work

The evidence, not the adjectives.

Every project here is public — live, or with its source open. Each diagram shows how the system is put together: what's plain code, what's a model, and where the guardrails sit.
  • Deterministic code
  • Model output
  • Guardrail
  • Data
  1. Sheet 01QueueStorm Investigator2026Hackathon build

    A support copilot for digital finance that can't make promises it shouldn't.

    Investigates customer complaints against transaction history and returns a structured, routable decision. Rules decide; the model only writes.

    • Python
    • FastAPI
    • Pydantic v2
    • OpenRouter
    • Gemini 2.5 Flash
    • pytest
    Read the case study Source on GitHub
    1. Ticket + transactionsIn / out
      Bangla, English or Banglish
    2. Request validationGuard
      Pydantic v2 schemas
    3. Reasoning engineCode
      Match, dedupe, route, set severity
    4. LLM draftingModel
      Agent summary + customer reply only
    5. Safety post-processorGuard
      No PIN/OTP asks, no refund promises
    6. Structured decisionIn / out
      Verdict, department, human-review flag
  2. Sheet 02Axo2026Live

    An AI product architect that turns an idea into a blueprint and build-ready prompts.

    Describe a product in plain language; get a structured blueprint — features, user flows, stack, architecture — and prompts tuned for Lovable, Claude, Cursor and Bolt.

    • LLMs
    • Prompt engineering
    • Structured generation
    • Agentic workflow
    Read the case study heyaxo.dev
    1. Product ideaIn / out
      Free text, any level of detail
    2. BlueprintModel
      Features, user flows, stack, architecture
    3. Saved blueprintData
      The reviewable source of truth
    4. Builder promptsModel
      Generated from the blueprint
    5. ExportIn / out
      Lovable · Claude · Cursor · Bolt
  3. Sheet 03SellWise2026Live

    Forecasts and customer intelligence for small online shops.

    A business-intelligence dashboard with three ML modules — sales forecasting, customer segmentation and churn prediction — served from an independent Python microservice.

    • React
    • Node.js
    • Express
    • PostgreSQL
    • FastAPI
    • Prophet
    Read the case study Live app
    1. Orders, customers, stockData
      PostgreSQL · CSV import
    2. Node.js REST APICode
      Orders, inventory, invoices, reports
    3. Demand forecastModel
      Prophet · 7 / 14 / 30 days
    4. Segments & churnModel
      RFM · logistic regression
    5. DashboardIn / out
      Forecasts, customer risk, stock alerts
  4. Sheet 04BUBT AI Assistant2025Open source

    A university chatbot that answers from official documents — and cites them.

    Retrieval-augmented generation over university PDFs: upload a handbook, ask a question, get an answer with the source documents listed.

    • TypeScript
    • Node.js
    • Express
    • React
    • Gemini
    • FAISS
    Read the case study Source on GitHub
    1. University PDFsIn / out
      Handbooks, admission guides
    2. Parse & chunkCode
      1,000 chars · 200 overlap
    3. EmbedModel
      Gemini text-embedding-004
    4. FAISS indexData
      Cosine similarity · persisted
    5. Grounded answerModel
      Top-5 chunks → Gemini
    6. Answer + sourcesIn / out
      Cited documents

Also built

Smaller builds and experiments.

Next step

Have a workflow that should run itself?

Tell me what it is, what data it touches and what “done” looks like. You'll get a straight answer on whether AI is the right tool — and what it would take.

Start a project

Free 30-min intro callReplies within 24h