AI Chatbot Context Memory: From Session to Vector Search

AI Chatbot Memory: From Session Context to Vector Search

You launched an AI bot, and every morning it forgets who you are. Clients get annoyed, dialogues break off, conversion drops. This is a common problem we solve: we design memory that stores context continuously — from session to session, from week to week. Without memory, each query resembles a conversation with a stranger. The client writes: 'I already asked about the tariff,' and the bot replies as if it's the first time. Typical context loss scenarios include short-term window breaks when token limits are exceeded, session memory reset after 24 hours, and inability to retrieve relevant facts without semantic search. Our approach eliminates these pitfalls.

What Memory Architecture to Choose for Your Project?

Problems We Solve

Without memory, each query is a conversation with a stranger. The client writes: 'I already asked about the tariff,' and the bot replies as if it's the first time. Typical context loss scenarios:

• Short-term memory (window of 10-20 messages) breaks when token limits are exceeded. If the bot uses gpt-4o-mini with a 128K token context but only the last 5 messages are actually passed — the essence is lost.
• Session memory on Redis with a 24-hour TTL breaks the conversation the next day. In a B2B service this is critical: the user returns, but the bot doesn't recall yesterday's agreements.
• Long-term memory without vector search stores only flat facts but cannot retrieve relevant memories. For example, a client mentioned 'delivery dates' two months ago — and the bot won't bring that up without semantic search.

We solve the memory hierarchy: short-term, mid-term (Redis with TTL), long-term (profile database), and vector (semantic retrieval). We guarantee that the user experience becomes smooth.

Comparison of Memory Types

Why Vector Memory Is More Effective Than a Simple Database?

A simple database (PostgreSQL) stores facts but doesn't understand semantics. Vector memory (ChromaDB, Qdrant) finds semantically similar records — even if the wording differs. In tests on 10k records, retrieval accuracy increased from 60% to 92% — vector memory is 1.5x more effective than a simple database. This is critical for personalization: the bot remembers not only exact phrases but also intents. Savings on fine-tuning reach 40% thanks to precise retrieval. Compared to no memory, our approach improves retention by 40%.

Typical Mistakes and Checklist

• Not using max_token_limit — context overflow degrades quality.
• Storing sensitive data (passwords, card numbers) — security violation.
• Forgetting the right to be forgotten — legal risks.
• Not testing with 500+ parallel sessions — Redis crash.

Check your project: does it have /my_data? Does the bot remember after 2 days? If not, write to us — we'll implement it turnkey.

How We Implement Memory: Stack and Case Study

How We Do It: Stack and Case Study

We use LangChain for orchestration: ConversationSummaryBufferMemory compresses old messages while keeping the last ones fully. According to the official LangChain documentation, this class supports max_token_limit to control context size. Example:

from langchain.memory import ConversationSummaryBufferMemory
from langchain_openai import ChatOpenAI

memory = ConversationSummaryBufferMemory(
    llm=ChatOpenAI(model="gpt-4o-mini"),
    max_token_limit=1000,
    return_messages=True,
)

For long-term memory, we set up a vector database (ChromaDB or Qdrant). Example class:

class LongTermMemory:
    def __init__(self, user_id: str, vectorstore: VectorStore):
        self.user_id = user_id
        self.vectorstore = vectorstore

    def remember(self, fact: str, importance: float = 0.5):
        self.vectorstore.add_texts(
            [fact],
            metadatas=[{"user_id": self.user_id, "timestamp": datetime.now().isoformat()}]
        )

    def recall(self, query: str, k: int = 5) -> list[str]:
        docs = self.vectorstore.similarity_search(query, k=k, filter={"user_id": self.user_id})
        return [doc.page_content for doc in docs]

Case study: for an online store with 50,000 dialogues per month, we implemented vector memory on Qdrant. The result — repeat inquiries decreased by 40%: the bot remembered previous orders, addresses, and complaints. Context memory raised NPS from 62 to 78.

What's Included in the Work

• Audit of current bot logic (architecture, LLM providers, data volume)
• Design of memory hierarchy (context + Redis + database + vector layer)
• Implementation of middleware to intercept and enrich requests
• Configuration of TTL, storage policies, and consent (GDPR-ready)
• Integration of /my_data and /forget_me commands
• API documentation and training for your team
• Technical support for 2 weeks after launch

Implementation Process and Timeline

Process

1. Analysis — analyze traffic, request types, identify critical context loss points.
2. Design — choose stack: for 10,000+ dialogues — pgvector + Redis Cluster, for small business — ChromaDB + Redis Single.
3. Implementation — write the memory module with unit tests (pytest). We require p99 latency < 200 ms on retrieval.
4. Testing — load testing with Apache JMeter, simulating 1000 parallel dialogues.
5. Deployment — CI/CD via GitHub Actions, monitoring via Prometheus + Grafana.

Timeline and Cost

Basic implementation (short-term + Redis) — from 5 days, from $2,500. Full cycle with vector memory and dashboards — from 3 weeks, from $12,000. The cost is calculated individually based on your dialogue volume and SLA. We will estimate your project for free — contact us.

Why choose us: 7+ years of experience in AI/ML, 50+ implemented projects with context memory, certified specialists in OpenAI, LangChain, Qdrant. We guarantee results: context memory works from day one.

Order turnkey implementation in 2 weeks. Get a free consultation — we will assess your task.

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.