ML Research Engineer · Bangalore, India

Mohammed
Yaseen Maniyar.

Research Profile
  • Experience: 6 years in ML and NLP across startup and research-oriented environments.
  • Focus: deep RL and LLM post-training, with an emphasis on state representation in RLVR.
  • Current thesis: history-as-state imposes a structural capability ceiling for reasoning policies.
  • Current project: latent Markov belief state with VAE dynamics and exploration via uncertainty.
  • Trajectory: PhD-oriented path targeting UCL within 2 years, toward uncertainty-aware world models for abstract reasoning.
in progress — apr 2026
Latent Markov States for RL Post-Training

Current RLVR defines state as unbounded token history — a degenerate MDP formulation that prevents RL from discovering genuinely new reasoning strategies. Motivated by the NeurIPS 2025 oral showing this capability ceiling is structural, not a data or compute problem. Replacing history-as-state with a VAE-based latent belief over reasoning trajectories: a compact, learned Markov state that captures where the agent is in the solution space, not what it said. Latent variance used as an intrinsic exploration bonus. Ablating against standard GRPO and the Markov state paper (arXiv 2603.19987).

RL post-training RLVR VAE GRPO capability ceiling Qwen2.5-1.5B
github.com/MohammedYaseen97/latent-markov-world-model ↗

Latent world models for abstract problem-solving — an agent maintaining a diffusion-based belief over solution-space position, navigated by RL-driven epistemic uncertainty reduction. As distinct from physical-world models (JEPA, LeWM), this targets the epistemic state of a reasoning agent. The VAE project is the tractable proof-of-concept; the diffusion model is the PhD.

01
GPU Scheduling with PPO

Gymnasium GPU scheduling benchmark on a discrete-event simulator with realistic bursty workloads and hard feasibility constraints. PPO end-to-end with action masking. Controlled ablations on two research contributions: cross-attention policy vs MLP, and lookahead representation design (off / per_gpu / sorted / CDF). Rigorous eval hygiene — deterministic eval, best-checkpoint selection, separate VAL vs held-out REPORT seeds.

Attn + CDF → REPORT 0.615 vs SJF 0.529 (~16% gain)
PPO Gymnasium action masking cross-attention
02
Perplexity Can Lie: Indic GPT-2

Multilingual GPT-2 from scratch — BPE tokenizer, pretraining on 40GB Indic text, SFT, and DPO from primitives. Key finding: raw perplexity is an invalid cross-tokenizer metric. Bits-per-byte normalisation reverses the ranking entirely. 4.6× compression improvement on Kannada, 2.75× on Devanagari vs tiktoken. DPO increased chosen log-prob +6.1%, decreased rejected −16.8%.

All 4 pipeline stages published on HuggingFace
BPE tokenizer pretraining DPO Indic NLP
MODELS
03
BPE Tokenizer Era

Standalone exploration of Byte Pair Encoding tokenizer design with Indic-aware Unicode split patterns. Isolated study of how tokenizer choices affect downstream compression and model performance — the findings that fed into the Indic GPT-2 project's core methodological contribution.

Compression and bits-per-byte analysis across Indic scripts
BPE Unicode tokenization
04
Idiom Understanding in Transformers

MSc dissertation. Empirical study of how Transformer encoders represent idiomatic meaning. Adapter-based transfer experiments probing cross-lingual idiom generalisation (English ↔ Portuguese). MirrorWiC-style fine-tuning pipeline on HuggingFace with BERT/mBERT baselines.

Evaluated idiom detection and cross-lingual transfer performance
NLP adapters BERT cross-lingual
mar–oct 2025
Featurely.ai
Bangalore
Machine Learning Research Engineer

Designed a cognitive model of human attention and decision-making grounded in Kahneman's System 1/2 framework; modelled decision processes as Reconfigurable Bayesian Belief Networks spanning domains, time periods, and psychographic profiles. Architected a research-quality benchmark evaluating synthetic humans against real humans across six behavioural dimensions at individual and population level.

jul 2024–feb 2025
Independent
Research
Independent ML Research

Multilingual GPT-2 from scratch; LLaVA clone (CLIP ViT + Phi-2); deep study of MoE, RoPE, KV caching, speculative decoding, flash attention.

mar–jul 2024
MightyBot.ai
Remote / Palo Alto
Senior ML Engineer

Planner Agent using Tree-of-Thought + LangGraph for multi-step tool routing. Hybrid search (Qdrant + BM25) over Google Drive integrated into RAG pipeline. Experiments across DSPY, Outlines, Instructor for prompt tuning and structured output.

mar 2023–feb 2024
Zevi.ai
Founding Team
NLP Engineer

Built and productionised a shopping assistant chatbot reaching #1 on Product Hunt. Reduced TTFT to 3–5s and cut operational costs ~80%. Hallucination detection, knowledge graph retrieval, transformer-based search pipeline.

dec 2021–dec 2022
LivePerson
Sheffield, UK
AI / Speech Research Intern

Speaker diarisation via novel attention-based clustering module. VQ-VAE and attention-CNN benchmarks on DIHARD challenge datasets.

RL & Training
PPO · GRPO · DPO
RLHF · action masking
reward shaping
Gymnasium
LLM & NLP
HuggingFace · TRL
PEFT · LoRA · SFT
LangChain · LangGraph
DSPY · vLLM
Infra
PyTorch · CUDA
Ray · Docker
Kubernetes · AWS
FastAPI · Python
sep 2021–sep 2022
Sheffield, UK
MSc Advanced Computer Science — University of Sheffield

Machine Learning, NLP, Text Processing, Parallel Computing / CUDA

aug 2015–jun 2019
Bangalore, India
BE Computer Science — PES University

Algorithms, Machine Learning, Artificial Intelligence, Big Data Analytics