AI Engineer · PhD Researcher, LSU · MSc Computer Science
AI systems grounded in real-world evidence.
I’m Asif Faisal Chowdhury, an AI engineer and PhD researcher at LSU. I build retrieval, multimodal sensing, and verification systems for industrial work.
Documents
Baton RougeAmerica/Chicago
Local time
Production AI, shipped
AI/NLP Engineer, Headless Technologies Limited. I built and deployed conversational AI, backend services, and APIs for engaze.ai.
- Platform scale
- 1,500+ stores
- Platform scale
- ~1M queries
Doctoral research
Stage 1 Manuscript
Measure behavior
94.5% mean balanced accuracy
Stage 2 Planned study
Validate AI interpretation
Stage 3 Planned study
Evaluate evidence-grounded coaching
Publications
8 further studies in manuscript form
- Automated Utility Bill Data Extraction and Analysis for Energy ManagementManuscript
- ConstructionQA: Evaluating Multimodal Large Language Models on Industrial Construction Drawing SetManuscript
- Benchmarking Retrieval Representations for Question Answering over Construction Drawing SetsManuscript · related version of “Local Multimodal Retrieval-Augmented Question Answering over Construction Drawing Sets”
- Large language models in industrial energy auditing: current state, architectural gaps, and the path to programmatic accuracyManuscript
- Local Multimodal Retrieval-Augmented Question Answering over Construction Drawing SetsManuscript
- Multi-Agent AI Framework for Objective Assessment of Excavator Operator Proficiency Through Temporal Behavior AnalysisManuscript
- Multimodal LLM Framework for Context-Aware Locomotion Prediction in Construction ExoskeletonsManuscript
- Skill Classification and Behavioral Analysis of Excavator Operators Using Wearable SensorsManuscript
- Trustworthy validation of industrial energy assessment reports: a neuro-symbolic validator and seeded-error benchmarkManuscript
Playground
Explore four interactive labs. Reported results stay separate from labeled synthetic demonstrations.
- Selective RAG risk–coverage
- AuditGuard deterministic vs. LLM verifier
- Bimanual joystick simultaneity
- Bayer dither and blueprint patch inspector
Selected engineering work
Executable verification, local retrieval, and wearable sensing. Read the contribution and evidence behind each system.
First author and system researcher
AuditGuard: verifying industrial energy reports
96.2% mutant-level exact-match
Designed the report-validation workflow and evaluated deterministic and model-assisted checking.
- Programmatic verification
- Energy auditing
- Evaluation
Co-author and retrieval researcher
Local multimodal drawing retrieval
65.8% Precision@1
Co-authored evaluation of local retrieval representations and fusion for construction drawing sets.
- Multimodal retrieval
- Local models
- Construction drawings
First author and doctoral researcher
Measuring excavator operator behavior
94.5% mean balanced accuracy
Analyzed joystick-mounted inertial measurements and separate gaze measures to study experience-related behavior.
- Wearable sensing
- Behavioral analysis
- Skill classification
Featured research & manuscripts
Selected findings, with the full research record one click away.
Manuscript
Trustworthy validation of industrial energy assessment reports: a neuro-symbolic validator and seeded-error benchmark
96.2% mutant-level exact-match
Reported finding · Abstract; Tables 4–6; Figure 1
Mutant-level exact-match
Highest reported value: Rules plus calculator, 96.2%.
Rules plus calculator96.2%Hybrid 8B43.0%Direct LLM 8B0.1%Percent
Data table
Mutant-level exact-match. 1,620 mutated reports in the full held-out evaluation. Condition Value Rules plus calculator 96.2% Hybrid 8B 43.0% Direct LLM 8B 0.1% Reported finding. Adapted from Trustworthy validation of industrial energy assessment reports: a neuro-symbolic validator and seeded-error benchmark, Abstract; Tables 4–6; Figure 1. Source and method
96.2% exact-match with rules and a unit-aware calculator.
Manuscript
Local Multimodal Retrieval-Augmented Question Answering over Construction Drawing Sets
65.8% Precision@1
Reported finding · Table 3; Section 4.2; Figure 11
Precision@1
Highest reported value: Triple fusion, 65.8%.
Raw sheet BM2545.0%Card BM2556.5%ColQwen259.0%Triple fusion65.8%Percent
- Whisker: 95% confidence interval
Data table
Precision@1. 407 human-authored questions; first retrieved sheet relevance. Condition Value Uncertainty Raw sheet BM25 45.0% 95% CI [40.0, 49.6] Card BM25 56.5% 95% CI [51.6, 61.2] ColQwen2 59.0% 95% CI [54.3, 63.6] Triple fusion 65.8% 95% CI [61.2, 70.3] Reported finding. Adapted from Local Multimodal Retrieval-Augmented Question Answering over Construction Drawing Sets, Table 3; Section 4.2; Figure 11. Source and method
65.8% Precision@1 by fusing text, sheet summaries, and visual retrieval.
Manuscript
Skill Classification and Behavioral Analysis of Excavator Operators Using Wearable Sensors
94.5% mean balanced accuracy
Reported finding · Table 4; Section 4.2.1; Figure 6; Table 5
Mean balanced accuracy, repeated five-fold validation
Highest reported value: Logistic regression, 94.5%. Whiskers show one standard deviation, not a confidence interval.
Logistic regression94.5%Linear discriminant analysis92.5%Radial-basis SVM92.2%Random forest89.0%Percent
- Whisker: plus or minus one SD, not a confidence interval
Whiskers beyond the scale end at the axis edge without a cap. Exact bounds are in the data table.
Data table
Mean balanced accuracy, repeated five-fold validation. Usable 36-trial IMU sample; twenty repeats of stratified five-fold cross-validation. Condition Value Uncertainty Logistic regression 94.5% SD 9.0 percentage points Linear discriminant analysis 92.5% SD 11.6 percentage points Radial-basis SVM 92.2% SD 10.7 percentage points Random forest 89.0% SD 12.2 percentage points Reported finding. Adapted from Skill Classification and Behavioral Analysis of Excavator Operators Using Wearable Sensors, Table 4; Section 4.2.1; Figure 6; Table 5. Source and method
94.5% mean balanced accuracy from wearable joystick signals.
Accepted Journal of Intelligent Construction
Exoskeleton locomotion mode prediction in construction using GPT-4o: Zero-shot learning from vision and speech
87.87% weighted F1, zero-shot GPT-4o
Reported finding · Article first page; Figure 6; evaluation protocol
Weighted F1
Highest reported value: CLIP, fine-tuned, 90.05%.
CLIP, fine-tuned90.05%GPT-4o, zero-shot87.87%ImageBind, fine-tuned78.84%Percent
Data table
Weighted F1. Article evaluation samples; sample splitting does not establish participant-held-out validation. Condition Value CLIP, fine-tuned 90.05% GPT-4o, zero-shot 87.87% ImageBind, fine-tuned 78.84% Reported finding. Adapted from Exoskeleton locomotion mode prediction in construction using GPT-4o: Zero-shot learning from vision and speech, Article first page; Figure 6; evaluation protocol. Source and method
87.87% weighted F1 from GPT-4o without task-specific training.
Manuscript
Multimodal LLM Framework for Context-Aware Locomotion Prediction in Construction Exoskeletons
94.3% reported classification accuracy
Reported finding · Table 1; Figure 3 caption; evaluation text
Reported classification accuracy
Reported classification accuracy, Representative run: 94.3%.
Representative run94.3%Percent
Data table
Reported classification accuracy. 33 correct predictions out of 35 command events in the representative run. Condition Value Representative run 94.3% Reported finding. Adapted from Multimodal LLM Framework for Context-Aware Locomotion Prediction in Construction Exoskeletons, Table 1; Figure 3 caption; evaluation text. Source and method
94.3% reported accuracy with local perception, API reasoning, and a clarification gate.
Engineering starts in the field
Energy Assessment Team Leader & Researcher
LSU Industrial Training & Assessment Center
September 2024 to present
I lead industrial energy assessments, from utility-bill analysis and on-site measurements to energy balances and client technical reports. The work connects physical measurements to checkable recommendations.
Assessment scope and team outcomesAuditGuard report verification
LSU Industrial Training & Assessment Center · September 2024 to present
Energy Assessment Team Leader & Researcher
I lead industrial energy assessments: utility-bill analysis, on-site measurement, energy-balance calculations, and client technical reports. Totals are team assessment outcomes.
Recommended annual electricity savings
8,208,899 kWh/year
16 assessments with recommendations
Implemented annual electricity savings
5,950,754 kWh/year
11 assessments with implementation reports
Open to a conversation.
AI engineering roles where systems meet real operations.
- Emailafschowdhury@gmail.com
- Phone+1 (225) 506-8796
- LinkedInin/afschowdhury
- GitHubgithub.com/afschowdhury
- Hugging Facehf.co/afschowdhury