Trajecta
A local-first AI agent with persistent memory, learned skills, tool orchestration, and evaluations.
View case study
HELLO, I'M
Md. Fahim Islam
I build AI systems that
retrieve, reason, and improve.
AI Engineer focused on LLM applications, retrieval, autonomous agents, memory, and evaluation.
Ask about my work
Questions about projects, experience, and research — answered from verified portfolio details.
Uses documented portfolio facts · Always verify important claims
FEATURED WORK
View all projects Engineering with evidence.
PhyRAG
Offline physics RAG with re-ranking, local models, and carefully measured retrieval quality.
View case studyQuestGen
Course-grounded Chat, Agent and Exam modes with human-approved question drafting.
View case studyClearCut
Model-efficiency research using structured N:M sparsity and low-rank SVD.
View case studyPANDA
Voice-enabled conversational screening support with evidence-based evaluation.
View case study↗Hierarchical Reasoning Embeddings in Hyperbolic SpaceResearch↗Traffic Accident Outcome PredictionML Pipeline
7Projects built end-to-end
0.86 F1RAG math QA · RAGChecker eval
0.82 faithfulnessGeneration quality on 4k QA pairs
83–94 %PANDA clinical item alignment
Experience & involvement
Research, then product.
Hands-on AI engineering across retrieval, conversational systems, and model compression — from research prototypes to deployed applications.
Experience
Machine Intelligence Lab · North South University
Research Assistant — LLMs & Representation Learning
May 2025 – Dec 2025- Designed hierarchy-aware text representations using hyperbolic (Poincaré ball) embeddings, learned token-attention pooling, and multi-segment refinement to support coarse-to-fine semantic retrieval.
- Built a conversational ADHD screening-support system that tracks symptom coverage across dialogue and uses a multi-model pipeline to extract evidence and produce structured assessment outputs.
Activities
Bear Summit · National Semiconductor Symposium 2025
Represented the Machine Intelligence Lab
2025AI, IoT & Robotics Day
National seminar · North South University
2025NSU ACM Student Chapter
General Member
2025Stack
Tools I reach for under the hood.
Retrieval, agents, and evaluation — from model internals to streaming interfaces.
Currently into
- Building LLM applications with retrieval, tools, and structured workflows
- RAG pipelines — retrieval, re-ranking, grounding, and evaluation
- Agent workflows for multi-step AI applications
- Efficient local LLM inference and model compression
- Multimodal AI — text, vision, and audio
Programming Languages
PythonTypeScript
LLM & Agentic AI
LangChainLangGraphLangChain Deep AgentsRAGTool CallingPrompt EngineeringGuardrails AIOllama
Machine Learning
PyTorchHugging Face TransformersLLM Fine-tuningModel Pruning
Retrieval & Data
QdrantFAISSSQLiteMySQLConvex
Backend & Frontend
FastAPIREST APIsServer-Sent EventsNext.jsReact
Systems & Infrastructure
DockerBifrost LLM GatewayGitLocal LLM Inference
Contact
Let's build something that remembers.
I'm open to AI engineering roles and research collaborations — happy to talk about RAG, agents, or anything in between.