You will get I Will Fine-Tune a Transformer Model for Your Text Classification Task
Project details
I fine-tune transformer models for text classification and back every result with a real evaluation, not just a training-accuracy number. My baseline reference project: a DistilBERT classifier fine-tuned on financial news text, reaching 0.9237 macro-F1 on a balanced held-out test set of 2,000 articles, against a 0.8718 macro-F1 TF-IDF + Logistic Regression baseline, a 5.2-point macro-F1 gain, evaluated per class, not just on average.
Every project includes a confusion matrix, per-class precision/recall/F1, and a written interpretation of where the model performs well and where it doesn't, so you understand the model's actual reliability, not just a single headline metric. Higher tiers include MLflow experiment tracking, source code, and a FastAPI endpoint so the model is callable immediately, not left as a saved checkpoint.
I hold an MSc in Applied Artificial Intelligence from the University of Warwick and build production ML systems end-to-end, from data through deployment.
Every project includes a confusion matrix, per-class precision/recall/F1, and a written interpretation of where the model performs well and where it doesn't, so you understand the model's actual reliability, not just a single headline metric. Higher tiers include MLflow experiment tracking, source code, and a FastAPI endpoint so the model is callable immediately, not left as a saved checkpoint.
I hold an MSc in Applied Artificial Intelligence from the University of Warwick and build production ML systems end-to-end, from data through deployment.
What's included
| Service Tiers |
Starter
$150
|
Standard
$300
|
Advanced
$550
|
|---|---|---|---|
| Delivery Time | 4 days | 6 days | 10 days |
Number of Revisions | 1 | 2 | 3 |
Number of Model Variations | 1 | 1 | 2 |
Number of Scenarios | 1 | 1 | 1 |
Number of Graphs/Charts | 0 | 2 | 3 |
Model Validation/Testing | |||
Model Documentation | - | ||
Data Source Connectivity | - | - | |
Source Code | - |
Optional add-ons
You can add these on the next page.
Additional Revision
+$30
Additional Model Variation
(+ 2 Days)
+$75Frequently asked questions
About Sourabha
Financial NLP Engineer | RAG Pipelines & FastAPI/MLflow Deployment
Bengaluru, India - 10:39 pm local time
Recent work: a DistilBERT classifier fine-tuned to 0.9237 macro-F1 on financial news classification, against a 0.8718 macro-F1 TF-IDF/Logistic Regression baseline. A RAG pipeline over 22 SEC EDGAR and FOMC filings, built with a Qdrant vector store and cross-encoder reranking, scoring 1.0 context recall and 0.93 context precision on RAGAS. A FastAPI service exposing both, backed by 183 passing tests and a Docker/GitHub Actions CI/CD pipeline, live on GitHub now.
I also design the agentic layer around these systems. For a cybersecurity assistant that placed 2nd at a University of Warwick hackathon, I built a LangGraph pipeline with a regex-based prompt-injection gate ahead of every LLM call and a groundedness-gated retry that checks citation-to-source overlap before returning an answer.
What I do best:
Fine-tuning transformer models for domain-specific text classification.
Building RAG pipelines: retrieval, reranking, and RAGAS evaluation.
Deploying models as FastAPI services with tests, CI/CD, and drift monitoring (PSI, Kolmogorov-Smirnov, Chi-Square).
Designing LangGraph agent pipelines with guardrails against prompt injection and hallucination.
I hold an MSc in Applied Artificial Intelligence from the University of Warwick (2026). I work well with clients who need a system that actually runs, not a notebook that stops at a results table — everything above shipped with tests and a working API.
Steps for completing your project
After purchasing the project, send requirements so Sourabha can start the project.
Delivery time starts when Sourabha receives requirements from you.
Sourabha works on your project following the steps below.
Revisions may occur after the delivery date.
Review Data & Confirm Scope
I check your dataset's structure and class balance, confirm what each label means, and agree the training plan before starting.
Preprocess & Fine-Tune Model
I clean and tokenize the text, then fine-tune a transformer model on your data, with every run tracked in MLflow.