Fiza Husain

I build machine learning systems end to end — data, models, serving, and the evaluation that proves they work. Most of what I do sits between research and production: reading the literature and running the experiments, then owning the result once it’s live and real users are hitting it. I’ve found the two halves are hard to separate, and that the interesting problems tend to appear where they meet.

Right now that means multimodal models. I’m the founding machine learning engineer at Stimuler, where I own the speech stack behind a conversational English-fluency platform — fine-tuning audio language models for recognition and speech-to-speech, and serving them fast enough to hold a real conversation. The users are non-native English speakers across India, Indonesia, and Latin America, which makes almost every published benchmark a poor guide; accented, conversational, noisy speech is exactly the regime where general-purpose models quietly fall apart.

Before this I was a Research Fellow at Microsoft M365 Research, working on AIOps — applying foundation models to incident management and to the Intelligent Monitoring problem for large cloud services, using operational telemetry at a scale that made most conventional approaches fall over. That work appeared at FSE and ICSE. Earlier still, at IIIT Hyderabad, I worked with Prof. Praveen Paruchuri and Prof. Sujit Gujar on privacy in deep reinforcement learning, published at AAAI. Five peer-reviewed papers, two as first author, and a patent.

The thread through all of it is measurement. It is usually harder than the modelling, and it is where I have been most wrong. A model that scores well on the metric everyone reports can be useless for the thing you’re actually building, and finding that out early is most of the job.

Outside of tech I draw and paint, and I remain devoted to cats and to mathematics.

selected work

Teaching ASR to keep the words that matter

Entity- and disfluency-aware speech recognition for accented conversational English. INTERSPEECH 2026.

How I evaluate a voice agent

What to measure in a speech-to-speech system when there is no reference transcript to compare against.

Serving speech models when the metrics lie

Why GPU utilisation is the wrong signal for multi-model inference pipelines, and what to instrument instead.

Building an LLM judge you can actually trust

Validating automated evaluation against human labels, and what to do when they disagree.

latest posts

Aug 24, 2026	How do you evaluate a conversation?
Aug 24, 2026	Curating training data with SQL, not annotators
Aug 24, 2026	Why word error rate is the wrong target

selected publications

Beyond WER: Entity and Disfluency Recall in Accented Conversational ASR

Fiza Husain, Ankit Pandey, and Yash Singh

In Proceedings of INTERSPEECH 2026 , 2026

Abs PDF

ASR systems optimised for Word Error Rate (WER) often miss named entities and filled pauses in accented conversational English, both critical for language-learning feedback. We present a three-stage pipeline for speakers from India, Indonesia, and Latin America: (1) heuristic SQL filters curating entity-rich training data at 2.8x the entity density of random sampling, (2) regional LoRA adapters fine-tuned on Qwen2.5-Omni-3B producing both verbatim and corrected transcripts in a single forward pass, and (3) a six-category error taxonomy validated by an LLM-based judge (83.8% agreement, 210 human-labelled samples). The pipeline achieves 80-85% entity recall (up from 53-55%), 76-86% filler recall (up from <5%), and 6-10% WER across 6k test utterances, outperforming Whisper and a commercial ASR on entity recall while matching a zero-shot 30B model with 10x fewer parameters. Paired bootstrap tests confirm that curation alone accounts for 2.8-4.2 pp of entity recall gain (p<0.0001).
Attention Enhanced Entity Recommendation for Intelligent Monitoring in Cloud Systems

Fiza Husain, Anson Bastos, Anjaly Parayil, Ayush Choure, and 3 more authors

In Proceedings of the 34th ACM International Conference on the Foundations of Software Engineering (Industry Track) , 2026

Abs arXiv

In this paper, we present DiRecGNN, an attention-enhanced entity recommendation framework for monitoring cloud services at Microsoft. We provide insights on the usefulness of this feature as perceived by the cloud service owners and lessons learned from deployment. Specifically, we introduce the problem of recommending the optimal subset of attributes (dimensions) that should be tracked by an automated watchdog (monitor) for cloud services. To begin, we construct the monitor heterogeneous graph at production-scale. The interaction dynamics of these entities are often characterized by limited structural and engagement information, resulting in inferior performance of state-of-the-art approaches. Moreover, traditional methods fail to capture the dependencies between entities spanning a long range due to their homophilic nature. Therefore, we propose an attention-enhanced entity ranking model inspired by transformer architectures. Our model utilizes a multi-head attention mechanism to focus on heterogeneous neighbors and their attributes, and further attends to paths sampled using random walks to capture long-range dependencies. We also employ multi-faceted loss functions to optimize for relevant recommendations while respecting the inherent sparsity of the data. Empirical evaluations demonstrate significant improvements over existing methods, with our model achieving a 43.1% increase in MRR. Furthermore, product teams who consumed these features perceive the feature as useful and rated it 4.5 out of 5.