Skip to content
View ForgedEmir's full-sized avatar
1 follower · 4 following
  • Belgium

Achievements

Achievement: YOLO

Achievements

Achievement: YOLO

Highlights

  • Pro

Block or report ForgedEmir

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.

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
ForgedEmir/README.md

Emir — Evidence Ledger

Evidence before confidence.

From applied ML and retrieval systems to evidence-first agents.
I’m currently exploring how memory, tools, skills and evaluation can make agentic systems more reliable.

Evidence ledger

System Public evidence Status
RAG Release, source code and documentation Implemented
neuro-spark Pipeline, tests and DVC workflow Academic project
diabetes-scan Notebooks, reports and evaluation Academic project
SourceLoom Architecture and decision records Design phase

Selected work

  • RAG — Hybrid retrieval for accounting-oriented documents, with SSE streaming, multi-user authentication, semantic caching and an MCP server.
  • neuro-spark — Distributed EEG motor-imagery processing with PySpark, frequency-band decomposition and MLlib classification.
  • diabetes-scan — Early-stage diabetes risk prediction using nested cross-validation, SHAP explanations and fairness evaluation.
  • CarWatch — Parallel Belgian used-car monitoring, deal scoring and Telegram alerts.
  • peppol-prospector — CBE and Peppol directory cross-referencing for Belgian company research.

Now investigating

SourceLoom — Adaptive Deep Research Engine is currently in its foundation phase. The research question is whether explicit evidence gaps can drive better follow-up queries than a naive multi-provider merge.

  • Evidence graphs and contradiction detection
  • Bounded agent loops with explicit stopping rules
  • Reproducible evaluation of memory, tools and skills
How I evaluate an experiment
  1. Establish a reproducible baseline.
  2. Define the budget, timeout and stopping conditions.
  3. Run a small validation before a full benchmark.
  4. Compare quality, cost and failure modes.
  5. Publish limitations; “unknown” remains a valid result.

Working principle

Test before claiming. Bound every loop. Treat every concrete number as a hypothesis until its source is verified.

Student in Belgium · Python · Applied AI · Agent evaluation

Pinned Loading

  1. RAG RAG Public

    Production-grade RAG pipeline for accounting professionals — hybrid retrieval (vector + BM25), SSE streaming, multi-user auth, semantic caching, MCP server. FastAPI + Qdrant + Cerebras.

    Python

  2. neuro-spark neuro-spark Public

    Distributed EEG signal processing pipeline — motor imagery decoding from 109 subjects using PySpark, frequency band decomposition & MLlib classification.

    Python

  3. diabetes-scan diabetes-scan Public

    Early Stage Diabetes Risk Prediction - classification supervisée avec nested cross-validation, SHAP, fairness. Projet ML HELMo 2026.

    Jupyter Notebook 1

  4. CarWatch CarWatch Public

    Real-time Belgian used car alert system. Scrapes 2ememain, 2dehands & AutoScout24 in parallel, scores deals mathematically, and pushes Telegram notifications instantly.

    Python

  5. peppol-prospector peppol-prospector Public

    Parse 2M Belgian companies from CBE, cross-references Peppol directory, identifies non-compliant businesses.

    Python

  6. SourceLoom SourceLoom Public

    Designing an adaptive deep-research engine for multi-provider retrieval, evidence graphs, contradiction detection and explicit stopping rules.