Tanveer Hossain Munim

Senior ML & Backend Engineer · Applied AI · Data-Intensive Systems

I build reliable AI products, distributed data platforms, and backend systems for high-stakes document, climate, and geospatial workflows.

5+ yearsSoftware delivery
1+ TB/dayData platform scale
100K usersPlatform growth
RemoteInternational experience

Engineering Profile

Systems, Products & Technical Leadership

I am a Senior Research Engineer at AI GeoLAB, where I build document-AI and cadastral systems for offline land-record digitisation. My work spans dataset design, model fine-tuning, evaluation, local inference services, review workflows, desktop packaging, and audit-ready data lineage.

Previously at RIMES, I was the sole architect and engineer for the backend, data pipeline, ML, and infrastructure of Timor-Leste's national meteorological platform. It processes more than 1 TB/day through a three-node Airflow cluster with 97.64% task success. I also built LLM and RAG systems, emergency-alert infrastructure, and the data layer behind Bangladesh's live 15-day flood-forecast platform.

Earlier, I led engineering at Interactive Cares as the platform grew from 8K to 100K users, and worked remotely for Shipday. I am comfortable owning systems end to end: architecture, implementation, deployment, performance, reliability, and technical leadership.

Engineering Focus

Applied AI Systems Backend Engineering LLM & RAG Systems Document AI Computer Vision Data Platforms GeoAI & Climate Reliable ML Evaluation
Tanveer Hossain Munim

Contact

Research Foundation

Selected Publications & Research

Work on reliable evaluation, uncertainty quantification, geospatial ML, and earlier human-centred computing.

  • 2026

    A Free Knob: Decoupling Calibration and Predictive Skill in Threshold-Based Evaluation

    Munim, T. H., Saiem, B. A., Sany, A.-A., & Hashem, T.

    ICLR 2027 submission · Datasets & Benchmarks

    A matched-calibration audit for thresholded rare-event benchmarks. On released CasCast checkpoints, the cascade-over-backbone CSI gap falls from 0.1601 to 0.0339 after monotone recalibration; across 450 SEVIR contrasts, 51 reverse sign.

  • 2026

    Counting People with Coverage: Distribution-Free Prediction Intervals for Crowd Counting via Crop-Consistency

    Munim, T. H., Ikram, F. Z., & Hassan, A. M.

    Preprint

    A post-hoc conformal wrapper for frozen crowd counters. Across four counters and four datasets, crop-consistency improves hardest-region coverage in all 12 model–dataset combinations; at matched coverage, intervals are 15% narrower than a label-trained uncertainty model and 29% narrower than isotonic recalibration.

  • 2026

    Anchored Risk Profiles: Exact Robustness Certificates for Fixed-Threshold Forecast Rankings under Acceptable Corrections

    Munim, T. H.

    In preparation · AISTATS 2027 target

    Exact certificates for whether forecast rankings survive every acceptable calibration-improving monotone correction.

  • 2026

    A Meteorological Analysis of the August 2024 Flooding of Eastern Bangladesh

    SPARRSO collaboration · author list in preparation

    In preparation

    A compound-mechanism analysis using ERA5, GPM IMERG, GOES, CMORPH, BWDB stations, and Sentinel-1 SAR.

  • 2026

    ShasthoSheba: Leveraging an mHealth Solution for Providing Healthcare Service to Orphanages

    Munim, T. H., Anan, M. T. T., Galib, F.-Z.-I., Saiem, B. A., & Islam, A. B. M. A. A.

    ACM Transactions on Computer-Human Interaction · Under review

  • 2024

    eDakterBari: A human-centered solution enabling online medical consultation and information dissemination for resource-constrained communities in Bangladesh

    Eliza, I. J., Urmi, M. A., Anan, M. T. T., Munim, T. H., Galib, F.-Z.-I., & Islam, A. B. M. A. A.

    Heliyon, 10(1), e23100

Invited Talks & Presentations

  • 29–30 April 2026

    “Climate–Agriculture Risk Modeling: Sri Lanka”

    Climate Services User Forum, South Asian Hydromet Forum (SAHF) · Malé, Maldives

    On behalf of RIMES

  • 9 October 2025

    “Machine Learning for Extreme-Event Detection in Agrometeorology”

    Capacity Building Training, National Center of Meteorology · UAE

    On behalf of RIMES

Selected Case Studies

AI, Backend & Data Systems

Production platforms, applied AI systems, offline products, and open-source engineering.

Khatian Platform — Offline Land-Record Workstation

An installable, pilot-ready offline workstation joining OCR, two-person review, audit history, searchable khatians, and cadex parcel processing. Built with Tauri 2, Rust, React 19, embedded PostgreSQL, and local FastAPI model sidecars; 30 of 30 unseen sheets completed with 14,536 parcels and no invalid geometries.

GCF CDIS — Timor-Leste National Meteorological Platform

Sole architect and engineer for backend, data pipeline, ML, and infrastructure. The platform processes more than 1 TB/day across five NWP models and two satellites using a three-node Airflow cluster, with 97.64% task and 96.20% DAG success. Cached NetCDF reads fell from 30 seconds to 80 milliseconds.

SAHF Regional Climate Intelligence — Grounded RAG

A provenance-aware decision-support system for nine South Asian countries. It combines live ENSO, IOD, MJO, SEAS5, and GloFAS data with primary-source-verified historical reports to generate structured sector advisories and grounded chat, with country/region attribution guardrails, source traceability, fail-soft refreshes, and bounded multi-turn context.

Khatian OCR — Bengali Land-Record Extraction

Fine-tuned Qwen2.5-VL-3B with QLoRA for two handwritten Bengali land-record schemas. Mouza-disjoint held-out tests reach 0.965 core-field F1 on 10,509 full khatians and 0.947 on 5,122 index records, with Bengali-aware evaluation and bootstrap confidence intervals.

cadex — Cadastral Plot Extraction

Turns scanned, georeferenced mouza maps into numbered plot polygons. A ConvNeXt-V2 U-Net, watershed polygonisation, an eight-model number-reader ensemble, and roster-aware assignment recover and correctly number 91.7% of plots across 22 held-out sheets.

NLAS — Multilingual Livestock Advisory

A beta hybrid-RAG assistant for Bangladesh's Department of Livestock Services. It runs Qwen2.5-7B Q4_K_M through llama.cpp on a 4 GB GPU, retrieves from 5,353 QA pairs, handles Bangla and Banglish, and records 71% satisfaction among its small beta user group.

wis2downloader — WMO Open-Source Contribution

Refactored WMO's downloader into distributed Celery workers with async queues and pre-enqueue spatial filtering, reducing ingestion latency from about four hours to 130 ms. Merged upstream and used in CDIS production.

CAP v1.2 Multi-Country Alert Infrastructure

Sole architect of the validator, Django RSS feed, MQTT, and n8n dissemination design used by DNMG Timor-Leste and BMD Bangladesh through the Grameenphone mass-dissemination integration. pycap-validator is co-authored with Anindita Bose.

Research & Evaluation

Selected Research Projects

Research that informs how I evaluate and ship reliable machine-learning systems.

FreeKnob Audit — Calibration vs Predictive Skill

A model-agnostic audit for threshold-based rare-event evaluation. It applies matched monotone calibration without retraining or spatial reordering, then reports the skill contrast that survives alongside pooled frequency bias.

Vocabulary Shift in Open-Vocabulary Segmentation

Ongoing research on when conformal guarantees for open-vocabulary segmentation break under vocabulary change, and how to predict the break before deployment.

SPARRSO Flood Intelligence System

A side project developing a satellite-based national flood monitoring and forecasting prototype. Its 90 m susceptibility model reaches national AUC 0.83 and beats the HAND baseline in all eight divisions; a Brahmaputra pilot reduces held-out daily-discharge error by 36.7% at one day.

Academic Foundation

Education

  • April 2018 – September 2023

    B.Sc. in Computer Science and Engineering

    Bangladesh University of Engineering and Technology (BUET)

    CGPA 3.62/4.00 · Undergraduate thesis: ShasthoSheba · Supervisor: Prof. A. B. M. Alim Al Islam · GRE 329 (Q170, V159) · IELTS 8.0

Engineering Toolkit

Technical Skills

Backend & Infrastructure

  • Python / Django / DRF / FastAPI
  • Rust / Tauri / TypeScript / React
  • Celery / multi-node Airflow / Redis
  • PostgreSQL / MQTT / SSE / uvicorn / nginx
  • Docker / RKE2 / Linux / HPC

ML Engineering & Evaluation

  • Conformal prediction (split, Mondrian, weighted, CRC)
  • Calibration & benchmark auditing
  • Bootstrap confidence intervals
  • Pre-registration & reproducibility
  • Bayesian modelling (PyMC)

LLMs, VLMs & Retrieval

  • Qwen2.5-VL / Qwen3-VL fine-tuning
  • QLoRA & full fine-tuning
  • llama.cpp / vLLM / Ollama
  • Hybrid RAG: BGE-M3, BM25, RRF, reranking
  • LangGraph / LangChain

Computer Vision & Document AI

  • OCR & key-information extraction
  • U-Net / ConvNeXt-V2 segmentation
  • Watershed polygonisation
  • Crowd counting
  • Open-vocabulary segmentation

Geospatial & Climate

  • Sentinel-1 / Earth Engine / HAND / GloFAS
  • ERA5 / IMERG / WeatherBench2 / SEVIR
  • xarray / NetCDF / COG / Zarr
  • GDAL / Rasterio / GeoPandas / PostGIS
  • martin / TiTiler / MapLibre / Deck.gl

Career

Professional Experience

  • April 2026 – Present

    Senior Research Engineer

    AI GeoLAB Ltd

    Building Khatian OCR, cadex, and the offline Khatian Platform for reliable Bengali land-record digitisation.

  • July 2024 – July 2026

    Senior Software Engineer

    Regional Integrated Multi-Hazard Early Warning System (RIMES)

    Sole architect and engineer for the backend, data pipeline, ML, and infrastructure of GCF Timor-Leste CDIS. Architected CAP v1.2 alerting, FFWC real-time data unification, and production ML systems including NLAS, Cirrus AI, and Heatshield.

  • June 2021 – August 2024

    Tech Lead (previously Backend Developer)

    Interactive Cares

    Scaled the platform from 8K to 100K users and daily active users from 2K to 16K; grew the engineering team from 2 to 8 and cut deployment time by 60%.

  • September 2023 – January 2024

    Software Engineer

    Shipday, Inc. (Remote)

    Built an AI assistant and an analytics dashboard that improved marketing-campaign ROI by 25%.

  • September 2022 – August 2023

    Data Engineer (Part-time)

    Survey of Bangladesh, Ministry of Defence — NSDI

    Automated geospatial processing by 80%, saving 2,400+ person-hours/year, and validated 50 GB+/month across 43 departments.

Product Leadership

Entrepreneurial Work

Founder, CEO & Lead Architect · Founded July 2025

MIRA AI

Multi-channel conversational sales agents for enterprise commerce.

Built with LangGraph, FastAPI, retrieval over 10K+ products, and Redis state. Designed for 100K end-users and currently serving four paying enterprise customers.

FastAPIHybrid RAGLangGraphRedis

Highlights

NVIDIA Inception Program · 2025
4 paying enterprise customers
Designed for 100K end-users

Credentials & Recognition

Awards & Affiliations

NST Fellowship Bangladesh National Science and Technology Fellowship · 2026
NVIDIA Inception MIRA AI · 2025
Accelerating Asia Winner Top 9 of 500+ startups · $100K · 2023
Bangladesh Mathematical Olympiad Champion · IMO team shortlist · 2016

Let's Connect

Get in Touch

I am open to remote AI/ML engineering and backend-heavy full-stack roles, especially teams building data-intensive or high-reliability products.

The quickest way to reach me is directly by email. I usually respond within 24 hours.

Email Me

Also find me on

Open To

AI/ML Engineering Backend Engineering Remote Roles Applied AI & GeoAI