banking77

NIRNAY 450M: open-weight calibrated decision model (Apache-2.0). Banking77 intent classification 87.9%, fitted ECE 0.045. Laya fork plus concept bottleneck, coarse-to-fine pointer and RLCD. Built by Eulogik.

Local Jev-style System One decision models on Apple Silicon Macs: AnyJev, Kev, Laya (Ollaya) and CLM benchmarked on English, Korean and BANKING77 (accuracy, latency, order-flip)

Local workbench for fine-tuning, evaluating, and serving LLMs on Apple Silicon with MLX-LM and FastAPI.

Banking intent classification using an MLP baseline and parameter-efficient LoRA fine-tuning of RoBERTa.

Llama-3.2-1B-Instruct, fine-tuned for the 77 Banking77 intents in one QLoRA run of about 25 minutes on a T4 with rsLoRA and NEFTune, scores 90.21% exact match on the test split and 90.28% once a trie over the label set constrains decoding. The README reports per-class accuracy and the confused pairs; the adapter is on Hugging Face.

Empirical study of candidate-order instability in non-autoregressive multi-candidate transformers, with cardinality scaling, permutation marginalization, cyclic shifts, calibration, and reproducible statistical evaluation.

LLM drift & quality monitoring: embedding drift, output-quality drift, and LLM-as-judge regression tracking

Verifiable handoffs between AI systems: Waggle packages decisions and evidence; Kea independently verifies, replays, or refuses them.

How many labelled examples a LoRA adapter needs to beat a frozen encoder: Banking77 learning curves on a laptop CPU

Banking support intent classifier with reproducible evaluation, batch predictions and human review.

Held-out intent benchmark on BANKING77 for few-shot intent classification, comparing TF-IDF, frozen MiniLM, and ProtoNet on unseen intents.

Small example of fine tuning ettin-encoder-17m over banking77 dataset on CPU only

I post-trained Nemotron Mini 4B into a CSR for Banks

An experimental NLP application comparing TF-IDF and DistilBERT on 77 BANKING77 intents, built with FastAPI, React and Docker.

End-to-end NLP system classifying banking customer queries into 77 intents. MLP baseline vs. RoBERTa fine-tuned with LoRA, with PII protection.

Frozen encoder + Mahalanobis prototype for class-incremental intent classification. 50+ experiments across BANKING77, CLINC150, HWU64, AG News. Matches fine-tuned baselines at 5MB state with zero forgetting, order-invariance, 455 QPS.

An example of agentic request routing and grounded knowledge lookup.

Benchmark and results for Closing the Compliance Gap: open-weights and proprietary LLMs for privacy-constrained contact center automation (under review, NeurIPS 2026 InfPriv workshop)

Fine-tunes Google FLAN-T5-base with LoRA (PEFT) on Banking77 to classify 77 banking intents. Trains only 1.77M params (0.71%) in ~2hrs on free Colab GPU. Served via FastAPI REST API with batch prediction, Docker support & evaluation metrics.