Research writing
Collaboration, compilation, provenance, and AI assistance that can validate its own LaTeX edits.
Independent research lab · Bangladesh
AI and software research from Bangladesh.
Shahriar Labs builds and studies software for collaborative LaTeX writing, code-generated STEM video, persistent agent memory, and programming in Bangla.
Primary systems
Three active research systems.
LetX covers scientific writing, QuantumSketch covers STEM video, and Context Heavy covers persistent agent memory.
Live↗
Collaborative research infrastructure
LetX
A real-time collaborative LaTeX editor with cloud compilation, Lexi, templates, and project history.
4,400+ registered users1,000+ compile-tested templates350+ universities and labsUnlimited collaborators on every plan
Active research↗
Generative scientific media
QuantumSketch
A prompt-to-video system that generates and renders ManimGL animations for STEM topics.
ManimGL renderingTemporal durable workflows
Active research↗
Persistent memory for AI agents
Context Heavy
A multi-tenant knowledge graph and hybrid retrieval system that stores agent context across sessions.
71.2% recall@10 on a 304-question LoCoMo sliceHybrid vector, text, graph, and reranking
Research directions
Scientific media
Code-generated animation and narration for explanations that need more structure than pixel generation provides.
Agent memory
Hybrid retrieval, knowledge graphs, temporal updates, and benchmarks for context that persists.
Accessible computing
Programming languages and learning environments built around local language and context.
Live evidence · September 25, 2026
Current LetX usage.
These figures come from the dated statistics page maintained by LetX.
Read the LetX dossier- 4,400+
- registered users
- 1,000+
- templates
- 350+
- universities & labs
- 60+
- countries
Lab journal
Recent research notes.
Systems · September 16, 2026
What a Personal WireGuard VPN Costs on AWS
Measured prices for a self-hosted WireGuard VPN on AWS, why Lightsail beats EC2 for this workload, and the user data trap that boots an instance with no VPN on it.
Evaluation · September 9, 2026
Testing LLM Applications Without Fooling Yourself
Most of an LLM app is ordinary deterministic code and should be tested as such. The model-dependent part needs evals, and evals have failure modes that produce confident, wrong numbers.
AI security · September 9, 2026
AI Guardrails That Hold: Architecture Over Prompts
You cannot instruct a model into being safe, because instructions and data are the same tokens. The controls that survive contact with an attacker sit outside the model entirely.
Contact
Discuss research tooling or agent infrastructure.
Contact us about research collaboration, technical questions, product feedback, or a partnership related to an active lab project.
Email Shahriar Labs